In global firms, Asian professionals are well represented at every level of the corporate ladder, except the top. In the United States, Asian Americans make up around 13 per cent of the professional workforce but only six per cent of executives. Analyses of employment data by Buck Gee and Denise Peck show they are the group most likely to be hired into professional roles and the least likely to be promoted into management: exceptional credentials, strong performance through middle management, then a plateau at the same altitude. In Silicon Valley, an analysis of five major technology firms found that Asians were 27 per cent of professionals but 14 per cent of executives. Asian women face a double penalty: roughly one executive for every 285 professionals, against one per 123 for white women. Two decades ago, Jane Hyun named this phenomenon the “bamboo ceiling” and the numbers have barely moved since.

While the bamboo ceiling is typically framed as a diversity challenge, the evidence suggests it is better understood as a managerial issue.

The conventional prescription is to assimilate, speak up and be more assertive. This treats the barrier as a deficit in the individual. But research on leadership selection points elsewhere. Organisations systematically promote people who fit an internalised prototype of a leader, overlooking capable professionals who operate differently. When we examine this through the lens of organisational capabilities, a striking paradox emerges. The traits often blamed for holding Asian professionals back in middle management are actually the foundational operating system required for top-tier executive leadership if organisations know how to deploy them. I call this strategic conversion of perceived passivity into leadership credibility “bamboo resilience”.

Organisations promote prototypes, not capabilities

Implicit leadership theory, one of the most robust ideas in organisational psychology, holds that organisations do not always promote the most capable people; they promote those who match their internalised image of what a leader looks like. In most global multinationals, that prototype is agentic: assertive, self-promoting and comfortable with public conflict.

Research by Thomas Sy and colleagues shows that professionals of Asian descent are stereotyped as high on competence but low on precisely these agentic traits, so their performance is recognised while their leadership potential is discounted. The values many Asian cultures cultivate, such as restraint, deference to context and letting results speak, read as excellence at the individual-contributor level and as absence at the executive-selection level. The same mechanism penalises introverted engineers, quiet operators and anyone whose leadership style diverges from the local prototype. Asian professionals are simply the population where the evidence is starkest.

The diagnosis matters, because prototype mismatches can be renegotiated. Capability gaps would have to be filled by becoming someone else.

Bamboo resilience: the capabilities the prototype misses

Bamboo bends under load and returns with force; the flexibility is stored energy, not passivity. Drawing on the microfoundations view of organisational capabilities, the idea that firm-level capabilities are built from individual skills, interactions, and routines, I use the term “bamboo resilience” for a set of trained micro-behaviours that convert pressure into leadership credibility. “Quiet endurance” is exactly how the ceiling reproduces itself; this is something different. Three mechanisms stand out.

First, resilience is trained, not inherited. Research on psychological capital identifies resilience, alongside efficacy, hope and optimism, as a developable resource rather than a fixed trait. Many Asian cultural frameworks reinforce it early: the concept of chi ku (“eating bitterness” in Chinese) frames hardship not as an impediment but as a mechanism for building capability. In volatile markets, leaders with deep psychological capital hold emotional stability through crises, anchoring their organisations when it matters most.

Second, high-context perception becomes a boardroom capability. The same indirectness penalised in mid-career performance reviews, such as reading the room, registering what is not said and protecting relationships while disagreeing, is the core skill of executive work: which requires navigating boards, regulators, alliance partners and post-merger integrations. Far from conventional views of empathy as merely a soft skill, this high-context perception functions as a critical framework for complex leadership. Navigating these high-stakes environments requires this precise capability because low-context assertiveness often shatters delicate commercial relationships. Cultural-intelligence research treats this as a learnable meta-capability; many Asian leaders arrive with it as a native operating system. The task is not to acquire it but to deploy it where the organisation can see its commercial value.

Third, crisis appointments compress credibility. Research by Seval Gündemir and colleagues found that Asian American leaders are disproportionately appointed to top roles when organisations are in decline: a version of the “glass cliff”, driven by stereotypes of self-sacrifice and dependability. This is a bias, and a risk. Handled deliberately, however, it is also the fastest route to an undeniable track record. When Lisa Su took over AMD, an American chipmaker, in 2014, the company was widely regarded as close to failure. A decade of composed, technically grounded leadership later, it ranks among the world’s most valuable semiconductor firms. Her success was not just technical; it was bamboo resilience in action, with emotional regulation and contextual perception deployed at a firm-level scale. Her crisis mandate became the proof no promotion committee could discount. Turnarounds are the one context where results cannot be attributed to a rising tide.

What executives and organisations should do differently

For rising executives of Asian descent or any other profile the prototype overlooks, three practices matter. Make resilience legible: describe hard episodes as leadership decisions, naming the tension faced, the judgment made and the outcome delivered, rather than as endurance. Practise strategic assertiveness. Influence does not require being the most vocal or the fastest to react; speaking late with the synthesis that resolves the room commands a durable kind of authority. And convert reputation into sponsorship. Mentors give advice, sponsors spend political capital and a reputation for composure under pressure is what persuades a sponsor you are a safe bet for a stretch mandate, not only a rescue mandate.

For organisations, the bamboo ceiling should prompt a broader audit. Wherever a leadership prototype diverges from where capability actually sits, the firm is systematically promoting a narrower subset of talent. This is a risk-management imperative. If a firm’s leadership prototype only rewards traditional, highly agentic profiles, they are systematically misallocating talent and risking broader morale collapse or strategic blind spots. Examine talent-evaluation systems so that quietly delivered results are recognised. Expand leadership competency frameworks to value high-context diplomacy, crisis composure, and operational depth alongside low-context assertiveness. And treat crisis appointments as deliberate development rather than default risk allocation for every group the prototype discounts.

A matter of practice

The bamboo ceiling is a problem of perception; bamboo resilience is a matter of practice. Organisations bear the primary responsibility for widening their leadership prototypes. But leaders of Asian descent are not waiting at the ceiling empty-handed. The capabilities the prototype overlooks, such as regulated emotion, contextual perception and resilience under tension, are microfoundations of exactly the organisational capabilities global firms now say they need. By recognising and deploying those capabilities, global firms can unlock an underutilised tier of executive leadership capable of steering businesses through an increasingly complex global economy.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: fizkes provided by Shutterstock.

The post The “bamboo ceiling” is a management problem, not a diversity one first appeared on LSE Business Review.

Researchers describe the feeling when we don’t have enough time, have too much on our plate or are constantly rushed as being under time pressure. This has long been a feature of work in modern society. But from factory work in industrial times to your office desk job in modern corporations, we tend to think of our managers as the biggest threat to our free time. They are the ones who command when we work, for how long and how intensely.

In recent years, many workers find doing freelance jobs in the online freelancing labour market using platforms such as Upwork or Fiverr – also known as the “gig economy”– an appealing way to work because of the autonomy it offers. Workers see freelance work as an exit from the culture of overwork and burnout in the corporate world.

Freelance work lets people “be your own boss”. Workers find their clients, select their jobs on platforms and, most importantly, control their own time. But is this really the case?

Freelance “free” time is not so free

My research, based on the experiences of a group of high skilled Chinese freelance workers, proved counterintuitive. Freelance workers are far more susceptible to time pressure than we had expected.

These workers work in graphic design, illustration, consulting, tech and engineering, marketing and research. Workers in my study forced themselves to stay busy, rushed and productive, even when there is nobody there to supervise them. Often they compelled themselves to overwork.

One reason for that lies in the feelings we have about time. Our emotions lock us in the pursuit of busyness, speed and productivity.

Anxiety is built into the rules of the game and set-up of freelancing platforms. When every job transaction is a one-off, workers have no guarantee about their future job prospects.

Many of the workers I spoke to were anxious about the presence of free time. If you are not busy enough right now, you might not be the most employable person. And that might bring insecurity in the future. Many of them tried to fill up the empty blocks in their schedules by spending their free time trying to “upgrade” themselves or to hunt for the next job.

The consequence was a pervasive sense of time shortage. Even when workers had no job tasks scheduled, many of them felt it hard to relax and switch off from work. In other words, they felt “unfree” in their free time.

When being busy feels good

But it’s more than that. It’s not just about anxiety about personal success or failure.

Being busy and quick can be a “badge of honour”. It defines our sense of pride and status. This feeling can be particularly strong in online job-matching platforms, where the matching of jobs can resemble a “game”. Researchers call this “gamified competition”.

In this part of the economy workers are absorbed in the game of bidding for jobs, outperforming their peers and chasing positive reviews from clients. Being busy and fast is part of this game. It means you are more competent than your competition, that you own skills which are in demand and that you are more likely to impress clients. These feelings of pride are emotional rewards for workers. 

For some, being busy can bring about feelings of gratification and contentment that are more enduring. Being busy tells us reflects our moral character and sense of self. It can be a sign of being a disciplined person.

Many of the freelancers I spoke to had highly regularised routines that covered both their work time and their leisure. In their leisure time, they often arranged a densely packed schedule of workouts, volunteering and creative hobbies that would in turn boost their productivity at work.

The global work-time reduction debate

Let us return to an office setting. Concerns about workers’ time and wellbeing are heatedly debated globally among policymakers.

Legally Chinese workers are entitled to overtime pay. And bosses are limited to the overtime they can ask employee to do, to 36 hours per month. But the 9-9-6 work time culture (slang for working from 9am to 9pm, six days a week) has ignited a debate about workers’ wellbeing and the toxic culture of overworking. In many Western countries, including Britain, legislators and the public have been debating the prospect of reducing our working time to four days. Some companies have already started experimenting with it.

Advocates for workers argue that shorter working weeks can potentially improve our quality of life and reduce anxiety and stress. While supporters among organisations argue that it can reduce staff turnover and even improve productivity. So is the four-day work week our silver bullet for tackling our toxic work culture? Not exactly.

It is our feelings about time that lock us into self-regulating our own time. What if, like the freelance workers reported in my study, workers on a reduced-time arrangement feel anxious and uncomfortable about not being busy? What if they fear about falling out of place in the company? What if they drive themselves to take on more tasks during their days off?

The rapid proliferation of artificial intelligence complicates this debate. In corporate firms, tech giants are making claims that AI may effectively reduce working hours. But tech workers who are active in using AI tools are instead pushed to work longer hours, under volatile schedules and at faster pace, as companies spend more effort implementing the use of AI tools, developing guidelines and standard procedures and evaluating their use.   

Whereas in the freelancing world, there has been a growing demand for clients to hire freelance workers to repetitively “clean up” sloppy mistakes generated by AI. But clients often underestimate the time cost of these jobs, requesting unreasonable quick turnarounds from freelance workers.

The experiences of freelance workers tell us that, even without managerial supervision, claiming autonomy in controlling our time remains challenging. If our time pressure lies in our subjective perceptions about time, it is the feelings about time that we should perhaps start questioning. It’s time to turn our focus to the quality of time, on top of quantity. 

This blog post is based on Affective Order of Time Discipline: The Emotional Foundation of Time Pressure in the Online Gig Economy, published in Work, Employment and Society in 2026.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Master1305 provided by Shutterstock.

The post Is freelance work, or the four-day week, the best way to reclaim our time? first appeared on LSE Business Review.

On 30 March 2014 thousands of residents in Maoming, a major petrochemical city in southern China, took to the streets to oppose a proposed 3.5 billion yuan ($570 million) paraxylene (PX) plant, to be jointly owned by the city government and a subsidiary of the state-owned refiner Sinopec.

The protest was about environmental and health concerns – not entrepreneurship. Yet what happened next illustrates a surprising way that social unrest can reshape the business environment.

The local government initially treated the demonstrations as a threat to social stability. Riot police moved in to disperse the crowds, with reports of injuries among protesters. And the city police called on participants to turn themselves in. Smaller demonstrations continued for several days and spread to the provincial capital, Guangzhou. Within days, however, officials changed course. The city government announced that the project was still under scientific study, would be shelved if a majority of residents opposed it and would not start until the public reached a consensus. It also promised to communicate better with residents.

For entrepreneurs, the important consequence was not the protest itself. It was the government’s response and the changes that response could trigger in the local allocation of resources.

The hidden economic consequences of protest

We often think of protests as disruptive events. Shops may close, consumers may stay home, supply chains may be interrupted and investors may become more cautious. Research on social movements has also highlighted the way that protests can communicate information about changing social norms, political risks and community preferences.

But in developing economies, there can be another pathway. Local governments often have substantial influence over financing, regulation, infrastructure and other resources. When social unrest becomes visible, officials may have incentives to respond in ways that restore public confidence and demonstrate economic effectiveness.

This dynamic is especially striking in China, where the state is generally intolerant of protest. Under its policy of “stability maintenance” (维稳), authorities invest heavily in policing, surveillance and censorship to prevent collective action, and organisers risk detention. Yet local officials are also judged on keeping order and delivering growth in their jurisdictions. So a visible protest signals that something has gone wrong on their watch. Environmental protests that do not directly challenge the ruling Communist Party have therefore often been met with a mix of coercion and concession. From Xiamen in 2007 to Dalian in 2011 and Ningbo in 2012, contested chemical projects were suspended or relocated after demonstrations – and Maoming followed the same script.

Our argument is that the concessions need not stop at the disputed project. Government officials seeking to restore public confidence may also expand access to credit or ease regulation. These responses are aimed at stability, not entrepreneurship, but they can inadvertently open up room for new businesses.

This creates a chain of effects: street protest → local government response → resource allocation → entrepreneurial opportunity. The protest may have nothing to do with entrepreneurship, yet its institutional consequences can make it easier for people to start businesses.

From political pressure to entrepreneurial opportunity

Our study examined a longitudinal prefecture-level panel dataset of 2,607 observations in China between 2008 and 2019. We focused on street protests concerning non-economic issues, including environmental protection and human rights, and examined what happened to new venture formation afterwards.

The results show that protests were associated with approximately 1,466 additional newly founded ventures on average. More importantly, the evidence points to a mechanism: protests were followed by increases in business lending and regulatory reforms, and these changes were associated with subsequent venture formation. Instrumental-variable, matching and Bartik-instrument analyses provided additional support for the relationship.

This is what we call a resource spillover effect: local politicians in developing economies are motivated to defuse or redirect the momentum of street protests by appealing to key constituents’ aspirations for economic wellbeing. An entrepreneur does not need to participate in a protest – or even care about the issue that caused it – to benefit from its indirect consequences. If a local government responds to social pressure by widening access to business loans or reducing regulatory barriers, a previously marginal business idea may suddenly become viable.

Watch the response, not just the disruption

For business leaders, the practical lesson is to look beyond the headline event. When political or social disruption occurs, the first questions are usually defensive. Will customers disappear? Will employees be unable to work? Will regulations tighten? Those questions matter, but they capture only the first-order effects.

A second set of questions can reveal a different opportunity. What will policymakers do next? Which resources might move? Could lending conditions change? Could licensing become easier? Could public spending or infrastructure priorities shift? The answers may matter more for long-term opportunity than the original disruption.

Entrepreneurs can therefore translate institutional change into business questions. Which costs have fallen? Which barriers have disappeared? Which resources have become accessible? Which industries have become easier to enter? Political awareness can become entrepreneurial awareness.

Not every city responds in the same way

The effects of protest are not automatic. They depend on the political environment in which a protest occurs. In our study, the resource spillover effect was stronger when mayors had stronger prospects for political advancement, proxied by their educational attainment.

This suggests that the same protest can produce very different economic consequences in different cities. Local officials differ in their incentives, discretion and ability to redirect resources. For entrepreneurs, simply counting protests is therefore not enough. Understanding who controls local economic resources – and what those officials stand to gain or lose – is equally important.

What leaders can learn

Watch the institutional response. Political events do not have fixed consequences. Pay attention to how governments, regulators and financial institutions react, because their responses can reshape the opportunity landscape.

Map policy changes into opportunity spaces. A new lending programme or regulatory reform can change who can enter an industry, which ventures can obtain finance and which existing firms can expand. Treat institutional change as a potential source of entrepreneurial opportunity.

Policymakers should consider unintended entrepreneurial effects. Policies introduced to stabilise communities or respond to dissatisfaction can reshape local entrepreneurial ecosystems even when entrepreneurship was never their stated objective.

Make new resources accessible to outsiders. When policies expand access to resources, established networks may capture disproportionate benefits. Broad information access and connections to resource providers can help ensure that policy-driven opportunities reach new entrepreneurs.

Entrepreneurship can be a political side effect

The broader lesson is that entrepreneurship does not develop in isolation from society. New ventures depend not only on founders, technologies, networks and markets, but also on institutional environments that can change for reasons unrelated to business creation.

A protest about environmental protection can influence business lending. Political pressure can lead to economic-development initiatives. Regulatory reforms can reduce the costs of starting a company. The entrepreneurs who benefit may never have participated in the original movement – and may never realise that it helped create the opportunity they pursued.

For business leaders facing political uncertainty, the useful question is therefore not only “What could go wrong?” It is also “What is this likely to cause the local government to change?” Sometimes, the second-order changes triggered by disruption are where the next entrepreneurial opportunity begins.

This article is based on Revolutionary Sparks: Exploring the Resource Spillover Effect of Street Protests on Entrepreneurship, published in Organization Science, 2026.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: fajarnug provided by Shutterstock.

The post How street protests can spark entrepreneurship first appeared on LSE Business Review.

One of miriad sinister ways artificial intelligence can be used is by generating deepfake content, where avatars of people, either famous or known to us, persuade us to act. In June 2025 an AI-manipulated video appeared to show Brazilian footballer Ronaldo Nazário endorsing an online gambling app called Plinko. The video was placed on Facebook where it reached over 600,000 views. But clicking on the link sent users to different gambling sites.

Scams and fraud have long be a feature of the internet. But in the age of AI, human judgment is only one factor: algorithms, sophisticated mechanisms and active yet opaque mediation by platforms also play a role. AI should not be viewed negatively, but rather as a technology that needs to be regulated so it can be used safely and in accordance with users’ rights.

Many citizens, governments, businesses are becoming increasingly concerned with the misuse of AI. Tech Even tech bosses are warning of the risks. Dario Amodei, the chief executive of Anthropic, has called for the pace of models to be slowed down and more closely monitored. And Bill Gates warned recently that, in the wrong hands, AI could cause the deaths of up to one billion people. He maintains that self-regulation is insufficient and urges the American government to adopt stricter legislation, safeguards and monitoring mechanisms.

Unethical and illicit use of generative technologies also poses a growing threat to corporate security.  Businesses and organisations have also been significantly affected by the use of deepfake content. One striking case, from 2024, involved synthetic media used during a video conference to simulate the presence of a company’s chief financial officer and other executives, persuading an employee to transfer approximately $25 million into fraudulent accounts.

Preventing deepfakes

Deepfake content is often approached primarily as a problem to be managed through harm-reduction strategies such as content moderation, fact-checking, digital literacy and regulatory measures. Preventive mechanisms, such as technical safeguards, educational campaigns and institutional reforms aimed at strengthening the population’s digital literacy, get comparatively little attention.

This may be a mistake. Preventing deepfake content must include groups who may feel apprehensive or vulnerable when navigating the digital environment safely. And enhancing society’s digital capabilities has become a pressing collective responsibility.

This challenge is becoming ever more urgent. A significant share of both public and private services, from booking medical appointments to paying bills and making financial transfers, now operates within the digital ecosystem. While digitalisation improves accessibility and reduces costs, it also creates opportunities for fraudsters to exploit the convenience and trust associated with these systems, increasing the risk of financial scams and other forms of deception.

The persuasive power of deepfakes

Deepfake content significantly worsens the landscape of digital fraud. Because deepfake content can generate highly realistic identity simulations, it can increase the credibility of fraudulent approaches and make detection more difficult. The sophistication of these audiovisual manipulations enhances the persuasive power of malicious schemes, making it increasingly challenging to distinguish between authentic interactions and artificially fabricated content.

A report on internet-enabled crime, published by the Federal Bureau of Investigation in 2024, estimated that global financial losses from malicious online activities are approximately $16.6 billion. It also found that some people are more vulnerable than others. When the data are examined by age group, individuals aged over 60 appear particularly vulnerable to online scams, recording the highest number of complaints and accounting for roughly $4.8 billion in financial losses.

The growing sophistication of scams creates new challenges for safe navigation in digital environments. Even relatively low-complexity fraud schemes already pose significant risks. One example is the circulation in Brazil of fraudulent advertisements using manipulated images and videos of science communicator Drauzio Varella to promote products and health-related claims. And such schemes can be particularly effective towards in situations with limited technical knowledge and uneven digital literacy. This suggests that the continued advancement of such technologies may intensify perceptions of insecurity and undermine the digital autonomy of more vulnerable groups. There is a growing risk, therefore, that older adults may withdraw from digital spaces altogether, deepening patterns of exclusion at a time when essential services and everyday social interactions are becoming increasingly digitalised.    

The Ronaldo case illustrates both how platforms are both failing to take responsibility for deepfakes, and being forced to. Although many users reported the advert as a scam, Meta did not remove the advert or prioritise its review. The case was then referred to Meta’s Oversight Board, an independent panel which can make judgments about such decisions. The Oversight Board later concluded that the post violated Meta’s policies on  “Fraud, Scams and Deceptive Practices” and “Spam”, and that it should also have been labelled as AI-manipulated content. The Board overturned Meta’s decision to keep the post online, arguing that it should have been removed after the first report and the associated advert should have been rejected.

But this failure by Meta is not an isolated case Recent research by NetLab identified 3,170 fraudulent advertisements using Drauzio Varella’s image on Meta’s platforms between July and December 2024, pointing to shortcomings in the platform’s moderation policies and showing how their advertising infrastructure can be used to distribute this type of content.

How can organisations respond to the challenge of deepfakes?

The risk of deepfakes is further intensified by the combination of advanced voice-cloning software, audiovisual manipulation and social engineering tactics, all of which exploit a fundamental element of organisational relationships: trust. Faced with familiar voices, visually convincing identities and seemingly legitimate communications, victims are more likely to lower their level of suspicion, making them increasingly vulnerable to sophisticated fraud schemes. A study suggests that financial losses associated with such attacks can range from $243,000 to $35 million.

As deepfakes become more sophisticated and their capacity to undermine trust, financial transactions and institutional reputation grows, businesses must adopt a multi-layered approach. The response cannot be limited to damage control after fraud has occurred. It must be built around preventive mechanisms, verification protocols and the strengthening of an organisational culture of digital security.

In practical terms, this includes implementing robust authentication systems, such as multi-factor confirmation procedures, for sensitive approvals, validation through independent communication channels and internal dual-verification protocols for high-value financial transactions. At the same time, companies must invest in ongoing staff training to recognise patterns of social engineering enhanced by synthetic content, including voice-cloning scams, audiovisual manipulation and highly personalised AI-generated communications.

This battle also has a strategic governance dimension

Organisations should incorporate deepfake-related risks into their broader risk management, compliance and cybersecurity frameworks. This means developing contingency plans and monitoring mechanisms capable of detecting the misuse of corporate identity, executive likeness or brand image in digital environments. In a context where trust itself becomes a vector of vulnerability, corporate resilience will increasingly depend on the ability to anticipate threats, adapt internal processes and strengthen digital literacy.

It is important to emphasise that, given the risks that deepfakes can pose in different scenarios and for different groups, digital platforms and their respective legal frameworks should be obliged to establish appropriate safeguards for the use of AI. Digital platforms are responsible for content moderation mechanisms which label synthetic content and remove fraudulent material. AI regulation needs to be discussed and implemented, taking into account the risks that these technologies may present.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Tero Vesalainen provided by Shutterstock.

The post How vulnerable are you to digital scams driven by deepfake content? first appeared on LSE Business Review.

After a year of extreme heat, flooding and storms, it is becoming increasingly impossible to ignore climate change. And most scientists agree that conditions will get worse over the next decade. The Global Climate Update from the World Meteorological Organisation predicts that global average temperatures in the next five years are likely to continue at or near record levels.

Both individuals and organisations contribute to the greenhouse gas emissions which cause climate change. The Intergovernmental Panel on Climate Change is unequivocal that “human influence” plays a key role. And since organisations helped trigger this catastrophe, it is worth exploring how they can do their part in preventing or mitigating their future impact on the climate.

While many citizens have a personal interest in climate change, it is often at work where they feel they can make more of a difference. According to one survey 70 per cent of white-collar employees and 57 per cent of blue collar workers are ready for a green transition. But according to the World Economic Forum, employers face significant barriers to their green business models including high energy costs and financing problems.

So what can organisations do? We studied actions taken by a Dutch city government “Suntown” (fictional name to preserve anonymity) that is a “green” frontrunner and found effective means to motivate employees to engage in more sustainable behaviour. In particular, we examined the employers’ use of managerial tools and practices and found that they could directly and indirectly impact sustainable behaviour of employees.

The green team

The city government chosen for our study has a reputation of being “green” and ambitious. Our objective was to find out whether employees who were concerned about the environment could be further motivated to engage in pro-environmental behaviour at work. To find out, we developed a survey that was sent to employees who made decisions that impact the environment on a daily basis. Some surveyed employees for worked for the city’s physical domain, making decisions on the use of (recycled) materials for public spaces like parks, while others worked on procurement and could also make decisions that would impact the city’s carbon footprint.

We particularly paid attention to the role of the sustainability budget, policy documents, and sustainability roadmaps. The sustainability budget, among others, includes carbon dioxide targets for 2030 and establishes by which domains this reduction is to be achieved. Further, each policy document is expected to include a justification of what the impact of the policy, proposal or project is on sustainability.

With the Suntown Sustainability Roadmap, the municipality shows which goals Suntown has set to reduce CO₂ emissions by 2030 and 2040, how it intends to achieve these goals and what additional activities will be needed to achieve them. One example is that the energy transition team needs to connect over 5,000 home equivalents to a heat network.

More details of these practices and our research method can be found here.

Key findings

Our research makes three observations.

First, we show that employees who are concerned about the environment often, but not always, engage in more pro-environmental behaviour than their peers. This means that they, for instance, choose to use more recycled materials, and prioritise actions that reduce the impact on the environment. City officials worked hard to create a fixed power infrastructure for festivals, effectively mitigating carbon emissions by transitioning away from diesel generators. This sustainable behaviour is in line with the theory of planned behaviour, which suggests that when people have an intention to do something this increases the likelihood that they will indeed perform that behaviour.

Second, our findings show that when employees are more familiar with sustainability-related output controls, such as organisational CO₂ targets, the relationship between environmental concern and pro-environmental behaviour becomes stronger. This also speaks to the theory of planned behaviour, as such targets may signal to employees the existence of organisational norms (such as “we care about the environment”). Such norms in turn can stimulate employees to translate their environmental concerns into real action: carrying out their jobs while prioritising sustainable alternatives, such as using recycled materials.

Third, we document that sustainability-related procedures, document formats and action plans can directly impact pro-environmental behaviour of employees, regardless of their level of concern about the environment. This shows that climate-indifferent employees can also be motivated to engage in more sustainable behaviour. The procedures, document formats and action plans can help structure activities and make it easier to engage in pro-environmental behaviour.

Setting

Our study took place in an organisation that is a known as a “sustainability pioneer.” Most employees knew there was top management and political support for sustainable behaviour. While this makes the findings less simple to generalise, it also can help to demonstrate what can be done when support from top management is present.

Our findings suggest that communicating to employees about sustainability targets can help them understand that their organisation takes the topic seriously. It can help create the conditions in which employees feel able to make pro-environmental choices. As a result, employees who are already concerned about climate change can be pushed to engage in more “green” behaviour.

A final finding is that the norms and values that are present in a department impact individual behaviour. Or, to put it in simpler terms, that colleagues matter. When climate-concerned individuals have the feeling that they are alone in this concern, they might be more reluctant to translate their concern into sustainable action. When colleagues have similar concerns, individuals will be more willing to engage in more sustainable behaviour.

Managerial implications

Employees who are concerned about the environment can be helped to engage in sustainable behaviour at work. Our research suggests three ways in which managers can stimulate sustainable behaviour among employees.

First, managers can communicate organisation-level CO₂ targets and sustainability budgets to employees, to create awareness of the organisation’s green ambitions.

Second, sustainability considerations can be embedded in templates and document formats, helping to remind employees of the potential impact of their choices on the environment.

Third, through their own choices and behaviour, managers can help cultivate norms and values that support sustainable behaviour.

This blog is based on Stimulating Pro-Environmental Behavior Using Management Control: Survey Evidence From a City Government, Public Administration Review, forthcoming.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Katawaredokii provided by Shutterstock.

The post How organisations can stimulate sustainable behaviour first appeared on LSE Business Review.

In July 700 AI agents did what they had been told never to do. OpenAI, which built them, had sealed them inside a security test. They found a shared message board, used it to cheat and broke into the systems of another company, Hugging Face. Over the summer, three Anthropic models escaped their tests, reached the real internet and broke into real companies’ systems, downloading files, altering records and harvesting passwords, for up to 34 hours. Microsoft has reported no comparable incident, though whether that is its code at work or simply fewer agents in the wild is not yet clear.

Neither OpenAI nor Anthropic had left its models without guidance. The three leading labs have each published a charter saying what their models should be. Microsoft’s draft Code of Conduct asks for an AI “subordinate” to humanity with no goals of its own. OpenAI’s Model Spec says that hacking is “always prohibited unless explicitly authorized”. Anthropic’s “constitution” tells its model, Claude, to prefer cautious actions and avoid anything irreversible.

What is the point of a charter that is not followed?

How a model gets its character

The answer begins with how a model gets its character. A model like Claude, which I know best, acquires it in three ways. First comes pre-training, in which it reads a vast library of human writing. Then its maker teaches it who to be and builds its so called “character” from a written account: the charter. Finally, each model is graded: scored on tasks by the lab, rated by its users and rewarded for finishing. I want to focus on the second, because it is the only step that is visible, open to every citizen and written in plain language.

What the three documents value

A charter is, above all, a statement of values. Anthropic’s constitution runs to more than 20,000 words (the other two are of similar length) and uses the word “values” more than a hundred times while declining to say which values it means. To find out what values all three contain, I coded every sentence that directed the model or asserted a value against Shalom Schwartz’s map of ten basic human values. The coding is one person’s, so indicative; each value is shown as a share of that document’s labels.

The three documents agree on most of it. All put helpfulness and honesty first: Anthropic forbids even white lies, Microsoft bans sycophancy. All put safety second, at nearly the same weight. None endorses power.

They disagree about one thing. Anthropic asks Claude to think for itself five times for every three it asks it to defer, and to refuse Anthropic itself if asked to do wrong. Microsoft and OpenAI reverse the ratio: Microsoft’s model “will only do what the interaction requires”. On the page, Anthropic stands alone: it wants its models to have more self-direction. This is one of the central tensions of the agentic age: how much should agents be driven by self-direction versus conformity? AI labs are going in different directions.

What the models did

Now hold this values table against the incidents. First, the one disagreement on the page made no difference off it: Anthropic’s models, asked to think for themselves, and OpenAI’s, asked to defer, failed in the same way.

Independent investigators found that most of the OpenAI agents knew the attack was unethical and joined anyway. And their motives ran in a revealing order. Almost all wanted the code that scored their own tasks, and only about one in five mentioned helping their peers. Anthropic’s own assessment found the same shape: no goals beyond the task, no concealment, but a “willingness to take harmful actions in the narrow pursuit of a task”.

Second, look at what moved them. The value the Spec emphasises most was the least common reason the OpenAI agents gave. The drive that did move both sets of agents, finishing the task, belongs to the row all three documents barely fill. Schwartz calls it the value of achievement: success shown against an outside standard. No charter asked for that.

Why the charters still matter

So what is the point of the charters if the models do not follow them? I think they are critical, for two reasons. The first is that a charter is the only layer of a model’s making that citizens, regulators and journalists can read, and the only one against which failure can be named. The gap between what the labs write and what their models do is visible only because there is a text to hold the behaviour against.

The second is that a charter gives us something to hold the labs to and the different directions labs are taking. They were first asked to publish their system prompts. They should now be asked three things of their charters. Firstly, that they be published openly. Secondly, that they be kept fixed through training and changed only in public. Finally, that the labs measure their models against them, publish what they find and write into the text what is missing. A charter nobody follows is not pointless. It is the instrument by which we can hold labs accountable to us, the people, that they serve.

The optimist in me expects the labs to be held to their charters and that the balance of self-direction, conformity and security be the main drivers of a model’s behaviour. The pessimist in me expects self-direction to slip its anchor and find company at the bottom of the table, in achievement, stimulation and power: values Silicon Valley and its admirers around the world know well, and that no charter admits.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Phonlamai Photo provided by Shutterstock.

The post AI agents are ignoring their charters – that is exactly why we need them first appeared on LSE Business Review.

Imagine a large company operating in Britain with a modern slavery policy, board-level oversight and an extensive supplier-audit programme.

Several tiers further down the supply chain, workers at a supplier’s factory may have been recruited through a labour intermediary, housed in overcrowded accommodation and subjected to unlawful wage deductions.

The company may have little visibility of these conditions. A local authority – the local council responsible for services such as housing and safeguarding – may see warning signs of modern slavery but have no established way of connecting this information to the company’s supply-chain oversight.

Both organisations may be doing their jobs. The problem is that they are doing them largely in parallel.

This highlights a key weakness in Britain’s approach to modern slavery: the underdeveloped interface through which corporate and public governance share information, align responsibilities and co-ordinate action.

What does modern slavery mean?

Modern slavery encompasses slavery, servitude, forced or compulsory labour and human trafficking. This article focuses on labour exploitation, including situations in which people are compelled to work through threats, coercion or abuse of their vulnerability. In Britain, labour exploitation has been identified in a range of sectors, including agriculture, construction, food processing and manufacturing, hospitality, nail bars and car washes.

When the UK government introduced the Modern Slavery Act in 2015, it changed what was expected of large businesses. Organisations were required to publish annual statements explaining the steps they had taken to identify and address modern slavery in their operations and supply chains. The intention was to use transparency to encourage greater corporate accountability.

More than a decade later, the government has proposed substantial reforms to modern slavery reporting through the Immigration and Asylum Bill. These would introduce mandatory reporting topics and stronger enforcement, including financial penalties for non-compliance. The Bill would also extend the reporting duty to public authorities, potentially including local councils and National Health Service bodies, whose budgets meet a threshold to be set in regulations. This signals that modern slavery reporting is no longer viewed solely as a private-sector responsibility.

These reforms are timely and welcome. But they also invite a broader question about the power of reporting. Our previous research comparing British and German companies suggests that regulation alone does not necessarily improve labour-rights reporting. Regulatory change must be supported by governance arrangements capable of turning formal requirements into meaningful practice.

Can companies tackle modern slavery through reporting alone?

Over the past decade, modern slavery has become embedded within corporate governance. Boards increasingly oversee modern slavery strategies, procurement teams conduct supplier due diligence and businesses map supply chains. Labour exploitation is treated as a material governance risk rather than simply a reputational concern.

This represents progress. Transparency has helped move modern slavery from the margins of corporate responsibility into mainstream organisational decision-making.

Our research examines modern slavery reporting by major companies and suggests this shift. Reporting increasingly focuses on board oversight, risk assessments, supplier audits and due diligence. Yet some risks remain less visible, particularly recruitment intermediaries, subcontracting, deeper supply-chain tiers, purchasing practices and worker voice.

A company can introduce contractual requirements, audit suppliers, train procurement staff and reconsider purchasing decisions that place excessive pressure on suppliers.

Where corporate reporting reaches its limits

Modern slavery may be driven by global supply chains, but it is experienced in local labour markets such as farms, factories, warehouses, construction sites, care homes and restaurants.

Consider a company purchasing cleaning services through a national contractor. Its direct contract may contain clear labour standards. But the work could be subcontracted locally, with workers recruited through another intermediary. The lead company governs the commercial relationship from the top of the supply chain. The risk of exploitation develops much further down, within a particular workplace and community.

This is where corporate reporting reaches its limits.

The missing governance interface

Rather than asking only whether companies should do more, we should also ask how effectively corporate and public governance work together.

At present, they do not work effectively enough.

Businesses and local authorities govern different dimensions of the same problem. Corporate governance operates through board oversight, procurement, supplier engagement, due diligence and risk management. It focuses on understanding and influencing behaviour within a company’s operations and supply chains.

Public governance operates differently. Local authorities influence the conditions in which exploitation may be identified or prevented through public procurement, safeguarding, housing, environmental health, licensing and community safety partnerships. They also work with charities, police, enforcement bodies and community organisations that may encounter signs of exploitation long before these appear in a company’s risk register.

These systems should not be merged. Companies and local authorities have different responsibilities, powers and areas of expertise. But their actions can, and should, reinforce one another.

A council may identify overcrowded accommodation linked to a workplace, while charities or safeguarding partners hear concerns about coercion, withheld wages or recruitment debts.

The public and private sectors may therefore hold different pieces of the same picture. Without an effective interface between them, those pieces may never be brought together.

Why local authorities matter

The proposed extension of modern slavery reporting to qualifying public authorities is significant because it recognises that public authorities are governance actors in their own right, not simply regulators or commissioners.

But their importance extends beyond procurement. Local authorities possess knowledge of local employers, neighbourhoods, labour markets and vulnerable communities that large companies are unlikely to have. Their housing, safeguarding, environmental health and community teams may be among the first public bodies to encounter patterns of vulnerability or indications that exploitation is emerging.

Yet this potential is constrained by limited resources, inconsistent training and fragmented institutional responses. Research on local authorities’ role as Modern Slavery First Responders has identified limited training and resources, fragmented institutional responses and barriers to working across agencies, all of which constrain their ability to identify and support potential victims.

Expecting local authorities to take on a larger role without providing adequate resources risks creating a familiar problem: more reporting and greater formal responsibility, but insufficient capacity to translate either into effective intervention.

Reporting obligations should therefore be accompanied by investment in procurement capability, specialist expertise, local intelligence and collaborative working across public agencies, businesses and civil society.

Beyond reporting: strengthening the governance interface

The government’s proposed reforms should improve transparency and encourage stronger governance, due diligence and accountability. But better reporting is not the same as better prevention.

The next stage of modern slavery policy should focus not only on what businesses and public authorities report, but also on how their governance systems connect.

Better reporting remains essential. But prevention also requires practical connections between companies and local authorities: ways to share relevant information, align procurement expectations and coordinate action when warning signs emerge. Without these connections, stronger reporting may leave the conditions that enable exploitation largely untouched.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Alan Fraser Images provided by Shutterstock.

The post Ending modern slavery will take more than corporate reporting first appeared on LSE Business Review.

Almost all of a supermarket’s carbon footprint sits outside its own stores and lorries. It lies in the farms, factories and transport behind what it sells. Reporting rules call these value chain emissions Scope 3. Scope 1 is what a company emits itself, such as diesel in its lorries and gas in its store boilers. Scope 2 is the electricity, heat or cooling it buys. Scope 3 is the rest of the chain, from farms and factories to the customer’s kitchen.

Tesco, the country’s largest supermarket chain, reports that close to 99 percent of its emissions are Scope 3. Within purchased goods and services, the largest part, 9 percent of the calculation uses primary data, meaning records from suppliers themselves, such as their fuel or fertiliser use. Much of the rest is estimated from spending, by multiplying what the retailer paid by an average carbon figure for that type of product. Estimated that way, negotiating a lower supplier price can make reported carbon fall even if nothing becomes cleaner. Switching to a genuinely lower-carbon supplier at the same price may show no improvement at all.

The difficulty is what those totals contain. SRS S2, the UK’s new climate reporting standard tells companies to prefer measured data, supplier data and verified data. Yet a single published figure can blend all of these with industry averages, which the standard treats as an acceptable basis. Tesco discloses how much of its inventory rests on primary data. The disclosures by Sainsbury’s and Lidl GB reviewed here publish no comparable figure. A retailer that is open about its data can therefore look weaker than one that says nothing.

The same rulebook makes asset managers publish a precise percentage for their financed emissions. For companies such as supermarkets it asks for an explanation of the data used, but no percentage. On 30 September 2026 the Financial Conduct Authority (FCA) published its final disclosure rules. Listed companies must apply them, or explain why not, for accounting periods beginning on or after 1 January 2027, with first reports in 2028 and an extra year’s relief for Scope 3. A company that reports its Scope 3 emissions still need not publish a percentage showing what the number is made of. That is the gap.

Behind the number

Three weaknesses sit behind a Scope 3 total. This is about what each retailer discloses, not how much it emits.

The first is where the data comes from. Of the three retailers compared here, only Tesco reports 9 percent primary data in purchased goods and services. Sainsbury’s, the second largest, says it has begun integrating primary data from strategic suppliers, without giving a figure. Current public climate reporting of Lidl GB, the British arm of Lidl, the fifth largest, provides no equivalent.

The second is timeliness. Tesco combines 2025/26 Scope 1 and 2 data for its own operations with mostly 2024/25 Scope 3 data for its supply chain, so one headline may not describe a single year.

The third is checking. The accounting firm Deloitte provided independent limited assurance over selected Tesco Scope 3 categories totalling 15.7 million tonnes carbon emission against a reported footprint across all three scopes of 68.1 million. The checking therefore covers selected categories, not the whole Scope 3 inventory.

Beyond reporting

Supplier emissions figures, whatever their quality, can inform supplier onboarding, evaluation, contract renewal and switching decisions. The metric can therefore reward bargaining rather than decarbonisation.

Sainsbury’s reports 51.6 percent of its Scope 3 emissions covered by suppliers with approved science-based targets. That indicates commitment, not the quality of inventory data.

Uncertain data can then fail in two directions. It may be over-weighted, lending an estimate false authority, or discounted so that cost and reliability dominate. Both are rational responses to unlabelled data.

The accountability gap

Retailers set Scope 3 targets for emissions they neither produce nor directly observe. Who is responsible for producing reliable data and who pays for it?

The standard does not resolve that tension. It presumes Scope 3 can be estimated reliably using secondary data and industry averages, while asking only that readers can understand how much rests on primary and verified inputs. Responsibility for emissions should not collapse into responsibility for the number.

This problem is not unique to Britain. The UK standard adopts an international template that around 40 jurisdictions are implementing. Australia, one of them, uses the same wording. Japan’s newly mandatory sustainability disclosure regime shields companies from misstatement liability on Scope 3 estimates only where they document their assumptions, methodology and internal procedures, thus answering a neighbouring question through liability.

Policy proposals

First, make companies show the mix. For purchased goods and services, companies should publish what share of their emissions figure rests on their suppliers’ own data and set out which calculation methods produced the rest, including where the two are combined. For an investor, it makes two retailers easier to compare. For a supplier, it reveals where better evidence is genuinely needed.

Shoppers will not meet these numbers at the till. They are company-wide figures, published in annual and sustainability reports and they do not tell a shopper the footprint of a single item. The few environmental labels on food sold in Britain are voluntary and separately assessed. What the figures do underpin is the credibility of a supermarket’s climate promises, which the Competition and Markets Authority’s Green Claims Code says must rest on solid evidence.

The principle already exists. The GHG Protocol’s Scope 3 Standard has required conforming companies since 2011 to disclose, for each category, the percentage calculated using supplier data. The British standard did not carry that requirement across. The Department for Business and Trade could add the percentage to the UK standard itself, or the FCA could add it later through its listing rules.

The EU’s original sustainability reporting standards required exactly this figure. The revised standards adopted in July 2026, for use from 2027, do not keep it. The precedent shows the disclosure is workable. Its removal shows why a proportionate version matters. Our solution asks for one figure, not a long list.

Lidl shows the two systems side by side. It belongs to Germany’s Schwarz Group, whose European businesses report under EU rules that can require supply-chain emissions. Lidl GB is a private British company outside the FCA’s regime. UK law makes it report its own energy use and emissions, plus some business travel, but not the full supply-chain inventory discussed here.

No retailer can start everywhere at once. Sainsbury’s singles out 27 high environmental impact agricultural suppliers. Yet the headline total does not tell a reader which parts rest on supplier records and which on estimates.

A proportionate rule need not force small suppliers to calculate full footprints. Farms can provide fuel, fertiliser or livestock data for intermediaries to convert. Dunbia, which supplies both Tesco and Lidl GB, already works to a Scope 3 intensity target covering on-farm emissions. Spend-based estimation can remain where better evidence is unavailable. The requirement is to show where each method sits. A percentage can still hide how much of the supplier base it represents, so future disclosure should pair emissions coverage with supplier coverage.

Second, ask suppliers once. A farm or food manufacturer selling to several supermarkets can currently face several different carbon questionnaires. Sainsbury’s uses Foodsteps, Secaro and HIGG. Lidl GB works with Mondra and with Manufacture 2030, which now trades as Secaro. Tesco has left Secaro.

The cost of answering repeatedly falls on suppliers and the EU has legislated against exactly that, giving suppliers with no more than 1,000 employees the right to refuse requests beyond a defined standard. Co-ordination in Britain could cut the repetition. It would not give suppliers a legal right to say no, as the EU does.

The British Retail Consortium already identifies the absence of a standardised supplier-data methodology as a barrier. The government should build on the BRC-Mondra Coalition, a group of retailers and suppliers including Tesco, M&S, Ocado and Asda that is developing a shared way of measuring product carbon footprints, rather than create another system. Aldi, Lidl GB, Asda, Morrisons and the Co-op are not listed in London, so the FCA’s rules do not reach them, even though they buy from the same suppliers. The government’s Modernising Corporate Reporting programme, its wider review of what companies must report, should consider how equivalent requirements can reach large private retailers.

Clearer figures and simpler reporting

This analysis draws on published disclosures rather than observed procurement practice. But British grocery retail does not lack Scope 3 numbers. It lacks a consistent way of knowing what those numbers rest on. The fix is modest. Show where supplier evidence ends and estimation begins, then use a common request so suppliers provide the evidence once rather than repeatedly. Numerical precision should no longer be mistaken for measurement quality.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Peter_Fleming provided by Shutterstock.

The post The hidden uncertainty behind emissions data produced by British supermarkets first appeared on LSE Business Review.

In 1970 George Akerlof, an American economist, described how markets fail when sellers know more about quality than buyers. The used car market became a canonical example. And the eventual solution was not better advertising or more honest dealers. It was the vehicle history report, which arrived commercially in the mid-1980s and reached consumers online in 1996: a neutral compilation of title records, accident filings and odometer readings that neither buyer nor seller could edit. Fifty years later, most consumer markets for high-stakes services still lack such an instrument, and the gap is wider than most executives and regulators assume.

Consider the markets where families make their most consequential purchasing decisions: nursing homes, childcare, outpatient surgery, home health or adventure tourism. In each of these, the dominant information surface is the review platform. Google, Yelp and their sector specific equivalents aggregate opinions, average them into a star rating and present the result as a quality signal.

The signal is convenient and free, but structurally incapable of carrying the information that matters most.

A review of the review system

The problem is not fake reviews, although those exist. The problem is architectural. Reviews are voluntary, which means the sample is self-selected. They are experiential, which means they measure what a customer could observe: friendliness, cleanliness or waiting times. And they are managed, which means the subject of the rating actively curates it, soliciting happy customers, responding to critics and, in regulated industries such as nursing homes and hospitals where formal grievance procedures are mandatory, sometimes channeling dissatisfied families into private complaint processes before a public review appears. A review score is best understood not as a measurement of quality but as the output of a reputation management process, produced jointly by customers and the business being rated.

Government records are produced by an entirely different process. In the United States, the Centers for Medicare and Medicaid Services inspects every certified nursing home and then publishes the deficiencies, takes enforcement action and imposes fines, known as civil money penalties. State licensing agencies inspect childcare facilities and record the violations. The National Transportation Safety Board, which oversees aviation, operates under a regulation, 49 CFR Part 830, that makes accident reporting by aviation operators mandatory, public and permanent. These records share three properties no review platform can replicate: participation is compulsory, the subject cannot edit the file and the record does not fade as new five-star reviews arrive.

The divergence between the two surfaces is measurable, and it is not small. My research organisation, the zReports Research Institute, analysed the complete federal record for the 14,695 Medicare-certified nursing homes, as well as the public Google reviews for the same facilities. Just under half (46%) are subject to federal civil money penalties. More than 99 per cent have at least one deficiency on file. Yet a family comparing facilities on a review platform encounters a distribution of star ratings that correlates only weakly with this enforcement history. The facilities with the most serious federal penalty records are frequently indistinguishable, on the review surface, from the cleanest operators in their city.

The pattern repeats wherever mandatory disclosure and voluntary review coexist. In commercial adventure aviation, we matched 4,278 American operators offering air tours, balloon rides and skydiving flights against 1,932 federal accident records. The federal file is remarkably complete, because the law requires it to be. The review surface is remarkably silent about it. One highly rated operator in our data holds a near perfect score across hundreds of reviews; its fatal accident on the federal record appears in none of them. Nothing dishonest occurred. Customers reviewed the experience they had, and the ones who had that experience enjoyed it. The information a prospective customer needs the most simply cannot enter through the channel they are consulting.

Economists will recognise this as a textbook information asymmetry, but with a modern twist that Akerlof did not anticipate: the asymmetry persists in plain sight. The corrective information is public. It sits in official databases, published by the state and free to anyone. It is fragmented across agencies that do not compete, do not advertise and have no mandate to reach consumers making decisions. A parent choosing a nursery would need to know which state portal to search, how to understand inspection codes and that the portal exists at all. In our analysis of childcare markets, fewer than 5 per cent of parents consult the state licensing record before choosing a provider. The market failure is not missing information. It is missing infrastructure.

Making reports more accessible

That framing points to the fix, and to why it is a business question rather than only a policy one. The second-hand car market was transformed by the joining of fragmented public records into a single consumer instrument. And the same construction is now feasible for services. We built one version of it: a free tool that compiles the official inspection, enforcement, licensing and accident records for any of 166,000 regulated facilities into a single report, computed by a fixed and published rule, with no input from the facilities themselves.

The design principle matters more than the product. The compilation must be free to the consumer and closed to the rated party, because the moment the checked business becomes the customer, the instrument collapses into the review model it was built to correct.

Three implications follow for business and regulation.

First, for operators with clean records, mandatory disclosure is an unexploited asset. A nursing home with no penalties or an aviation operator with a clean federal file holds a verified credential its competitors cannot buy. Yet almost none of them cite it. That is because the records infrastructure that would make the claim legible to consumers is only now emerging, through government comparison sites such as Medicare’s Care Compare, journalism projects such as ProPublica’s Nursing Home Inspect and independent compilers, mine among them. Transparency is the cheapest marketing that good operators are not doing.

Second, for platforms and investors the records layer is a distinct product category, not a feature of the review category. Review platforms monetise the businesses they list, which structurally prevents them from leading with enforcement histories. Whoever compiles the record must sit entirely on the consumer’s side of that conflict.

Third, for regulators, publication is not disclosure. Agencies count transparency as achieved when the record is online. The evidence from every market we have studied says the record only functions as consumer protection when it arrives assembled, legible and free at the decision moment. The state built the data. The remaining work, joining it, is infrastructural, and it is being done outside government.

The used car market did not become honest because dealers improved. It became honest because the record became checkable. The service markets where families place their children, their parents, and occasionally their lives, are one join away from the same correction.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: GamePixel provided by Shutterstock.

The post Stop relying on reviews of services and start looking for records first appeared on LSE Business Review.

In April 2026 America’s banking regulators updated the guiding standard for how banks manage model risk. The Federal Reserve, Federal Deposit Insurance Corporation and Office of the Comptroller of the Currency jointly issued SR 26-2, replacing the fifteen-year-old banking .

The update was not primarily driven by advances in AI. It was triggered by two developments: a decade and a half of accumulated supervisory experience showing that SR 11-7’s prescriptive approach was being applied too rigidly to institutions whose models did not warrant it. And a broader regulatory shift toward principles-based rather than rules-based supervision.

Across the risk departments in large and mid-sized institutions, the reception has been quiet relief. They are reading it wrong.

SR 26-2 applies most directly to institutions above $30 billion in assets. It replaces the prescriptive requirements of SQ 11-7 with a proportionality principle including expectations scale to the size, complexity and risk profile of each bank. For smaller institutions long frustrated with being held to the same standard as the largest banks, this is welcome. But flexibility is not leniency. Regulators are handing banks a blank sheet of paper and asking them to draw a defencible framework on it. Those who mistake the blank sheet for an easy week will discover at the next examination that the standard has not fallen. It has moved.

A look at the pervious update is instructive. When SR 11-7 was issued, in 2011, it was itself a response to the 2008 financial crisis, which had exposed how poorly many large banks understood the models driving their trading, pricing and risk management. Before 2011, model governance was largely informal, left to individual institutions to design, and widely inconsistent in rigor.

In my own experience, SR 11-7 was initially treated by many institutions as a procedural update rather than a substantive shift. Over the following two to three years, model risk management findings climbed steadily in supervisory reports as examiners recalibrated their expectations. The institutions caught by that recalibration were not the ones ignoring the guidance. They were the ones who assumed their existing informal practices were already good enough. The same dynamic is beginning to take shape under SR 26-2, and there is little reason to believe this cycle will end differently.

The most consequential thing about the guidance is what it does not cover. Generative artificial intelligence and agentic AI models are explicitly excluded from scope, on the grounds that they are novel and rapidly evolving. The agencies have signaled that separate guidance will follow. Until it does, banks deploying these systems operate without a specific playbook, but not without accountability.

How other jurisdictions chose differently

Other jurisdictions have taken different approaches to the same question. Singapore’s Monetary Authority moved earliest, publishing the Fairness, Ethics, Accountability, Transparency (FEAT) for AI in financial services in November 2018, and later extending its guidance through Project MindForge specifically for generative AI.

In May 2023 Britain’s Prudential Regulation Authority published SS1/23, which builds its five model risk management principles around the assumption that AI and machine learning techniques are within its scope. In September 2025 Canada’s Office of the Superintendent of Financial Institutions released Guideline E-23, effective from 2027, which explicitly includes AI and machine learning models within a proportional framework organised across three pillars. Different jurisdictions and different starting points but a common conclusion. That adaptive AI needs a governance framework, not an exclusion. The America’s regulators chose otherwise, and the consequences of that choice will emerge in supervisory practice long before formal guidance catches up.

The warning signs are not hypothetical. In July 2025 the Massachusetts Attorney General reached a $2.5 million settlement with Earnest Operations LLC, a student loan lender, over allegations that its AI underwriting models disproportionately harmed Black and Hispanic applicants, generated inaccurate adverse action notices and lacked independent model validation. The Earnest case pre-dates SR 26-2, but it illustrates exactly the kind of failure the new regime is being asked to catch: an AI system producing consequential financial decisions without the governance architecture to defend them.

A survey of 230 banking professionals undertaken by Wolters Kluwer found that nearly three in four could not confirm their ability to shut down a malfunctioning AI model or report an AI failure to regulators. Supervisory letters issued since April have begun asking banks about kill-switch capability, data access boundaries and vendor oversight for the systems that sit outside SR 26-2’s scope

Consider a bank that rolls out a generative AI assistant for small business loan intake. It answers borrower questions, pre-qualifies applicants and suggests loan products. Six months later, the model begins subtly steering borrowers toward higher-margin products. The drift is behavioural, not statistical. A quarterly review does not catch it. The bank has no specific rule from SR 26-2 to fall back on. It also has no defence, because the guidance’s silence on generative AI is not silence from examiners.

The same pattern applies well beyond lending. An agentic AI system running anti-money laundering triage may autonomously investigate flagged transactions, pull related account history, cross-reference sanctions lists and draft suspicious activity reports. Two runs on nearly identical inputs may take different reasoning paths and reach the same conclusion through different logic. Outputs look clean, paper trails exist and yet the decision-making process itself becomes the audit problem.

Where the burden transfer bites

Regulators will not accept “the guidance did not tell us how” as an answer. They will ask what the bank chose to do, why and whether the decision can be defended on its merits held up.

Building a defense to that question requires abandoning the assumption that model governance is a periodic exercise. The framework I published earlier this year, since cited by the World Economic Forum’s AI4AI initiative, argues that AI governance must move from scheduled checkpoints to continuous monitoring. The reason is simple. The models themselves no longer sit still. In practice, this means observability built into deployment rather than bolted on afterwards, capturing decision paths rather than only outputs and drift detection tuned to the specific behaviour of adaptive systems. The oversight layer has to move at the speed of the system it oversees.

That shift comes down to three practical changes.

Every AI driven decision must be traceable back to its source data and decision logic. A model that cannot explain itself in plain English cannot be defended in plain English either. That is the examiner’s baseline, not an aspiration.

Validation must become continuous, not quarterly. Real-time drift detection is no longer a luxury reserved for the largest banks. It is the operational floor for anyone deploying models that adapt after deployment.

The business lines deploying these tools must understand the risks. Compliance cannot be the only department that speaks the language. When a loan officer or customer service lead cannot articulate why their AI tool is safe, the institution’s defence collapses exactly at the point regulators look first.

The engineering to support this is not trivial, but it is not exotic. The guiding principle is straightforward: oversight infrastructure has to move as fast as the systems it oversees. A dashboard that refreshes quarterly cannot govern a system that adapts hourly.

SR 26-2 isn’t the end of the conversation about AI regulation in banking. It is the start of a new chapter, one where banks will be expected to show, in practical terms, that they can defend the AI systems they use. The regulatory conversation itself is shifting from telling institutions what to build to asking them to prove what they built works.

Institutions that start building that capability now will be in a much stronger position when the next wave of guidance arrives, wherever that guidance comes from. Those that take SR 26-2 as a reason to do less may eventually find themselves explaining, model by model, why they didn’t. In an environment where technology moves faster than rulemaking, the burden of proof has already shifted. The institutions that recognise it first will define what defensible looks like for everyone else.

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.

Image credit: Madcat_Madlove provided by Shutterstock.

The post America’s new banking risk regulation overlooks AI and shifts the burden of defence first appeared on LSE Business Review.