In recent weeks Bernie Sanders, Trump, and OpenAI have all aligned around the principle of the American public securing a financial value from Big Tech. The nature of these proposals vary from a one of 50% tax, through to public shareholdings, and public wealth funds.

Given how these models were trained, thinking about a public dividend is important. But we need to be conscious that the issue in Britain is different, and more hazardous than the situation Sanders is addressing in the US. Moreover, the window to do something about it is closing faster than most people realise.

The conversation about AI and jobs in this country remains stuck on a familiar loop: will the machines take our work, and if so, how do we retrain the people left behind? These are real questions. But they are the wrong starting point. Before we get to displacement, we need to talk about extraction.

Here's what's actually happening.

Foundation models - the AI systems commonly used at work to allow text, image and sounds to be generated - don't just perform tasks, they learn from them. Every time a worker uses an AI-integrated tool, they generate data about the work that is being done: in what sequence, with what judgments, by which methods. This isn’t about privacy, and personal data covered by existing laws. This is something far bigger, and more economically significant. It is the encoded know-how of the British workforce: decades of refined practice, institutional memory, and hard-won skill, being translated into training data.

Across various sectors – knowledge work, professional services, logistics, production – this know-how data is being collected at scale, and then funnelled into the same handful of American companies that dominate global digital infrastructure. Under the standard terms of many Foundation Model, AgenticAI and ‘Software as a Service’ (SaaS) contracts, there is nothing to stop this know-how from being used to train the next generation of models. This means that British workers are generating the data that will train the AI that will replace them. This value isn’t going to their employers, to funding hospitals, improving UK productivity statistics, or a new wave of British disruptor firms: the value from that process flows almost entirely elsewhere.

This is why Sanders’ idea sounds great - workers having a stake and reward from the AI transition - but won’t fly in Britain because we don’t have the privilege of these firms being national assets to tax. Rather than how to distribute the gains, the more immediate question is how to protect our own underlying assets: the knowledge and skill of the workforce from being quietly transferred overseas.  

In recent research at the Institute for the Future of Work, we found that most British firms simply don't know that this is happening. They are not reading the contracts they sign when adopting new solutions with this kind of data protection in mind – be that for their own sustainability, or the benefit of their employees. Further, it isn’t being disclosed in standard contract terms what data will and won’t be used for training models by third parties. That matters because, as one major multinational consumer goods firm told us, the parts of its business where AI tools were being considered - procurement, distribution, supply chain coordination - are the core of the business, the result of decades of operational learning. Handing this data over presented an ‘existential threat’ to their business model – quite aside from what it meant for workers.  

The potential for radically concentrated power, as know-how is accumulated by a small handful of providers, should not be underestimated. This data - the collective know-how of British workers, reflecting decades of public and private investment - must be recognised as a strategic asset and governed as such. Despite being held by workers within and across UK enterprise.  

Models are trained quickly and may soon be patentable: once the insight is captured, the learning can’t be taken back. We must act now.

Industry should wise up on the contractual and technical protections it needs to safeguard its commercial interests.

Workers and their representatives should be engaged in the conversation about how their work is being used for training, underpinning this significant value transfer.  

The government must begin to analyse these risks as they sit above the level of workers.  

The coming transition will require coordination and oversight in a systemic way.