Ferreras proposes three reforms spanning labs, workplaces and boards

Isabelle Ferreras used her November 19 commentary in The Guardian to argue that recent departures from AI safety roles reveal failures in corporate governance inside frontier laboratories. Ferreras is an accelerator fellow at the University of Oxford’s Institute for Ethics in AI, FNRS research director at the University of Louvain in Belgium and co-founder of DemocraticIntelligence.org. A photograph accompanying the commentary showed a protest outside Anthropic’s San Francisco headquarters on Tuesday, before the company’s IPO. The photograph was credited to Laure Andrillon/Reuters.

Ferreras said departures by safety researchers and engineers had become a recurring pattern over the past two years, with workers leaving after warning that commercial pressures were taking precedence over safety guardrails. She cited the OpenAI product-release safety-report overseer, who said last month that he was quitting the ChatGPT creator because the company’s culture was “broken” and the AI industry was too careless in developing the technology. The previous month, an Anthropic researcher quit after warning that AI “could kill us all by the end of the decade.”

Ferreras said society applauded the courage of those workers and debated their warnings for a news cycle before moving on. She said this missed the deeper institutional failure the resignations revealed: “exit is not voice.” She invoked political scientist Albert Hirschman for that maxim, arguing that a system requiring conscientious engineers to leave their jobs before raising safety concerns shows that corporate governance is broken. Each departure also removes institutional knowledge needed to steer AI systems safely, she said.

The commentary came as policymakers, tech workers, trade unionists, members of parliament and experts from global institutions including the Vatican and the UN gathered in London that week to discuss AI safety. Ferreras said governments cannot regulate AI safely from the outside if the companies building it remain autocracies on the inside. Her proposed strategy combines public regulation with worker power and democratic reforms inside corporate governance.

Ferreras said regulatory debates remain trapped between treating AI as a consumer product to be tested for defects and treating it as a geopolitical race to be won by relaxing rules on business. She cited Ursula von der Leyen’s EU digital omnibus, saying it delays implementation of the EU’s landmark 2024 AI Act and its requirements for developers of “high-risk” AI systems. Corporate lobbying, she argued, promotes a choice between safety and innovation by describing legislation as an excessive burden.

Within frontier laboratories such as OpenAI, Anthropic, Microsoft and Google, Ferreras said a handful of executives and capital investors decide which values are coded into models that shape millions of lives. She said exercising governing power in a democratic society requires the consent of the governed, yet corporate law still treats firms as the property of shareholders and labor investors, who put their intellect, health and lives into the firm, as subordinates. Public rules alone, she said, would not resolve that imbalance. Her proposed dual strategy combines solid public regulation from the outside with reforms that enter the corporate “black box” and align governance with democratic principles.

For AI development, Ferreras said governments must protect and empower those who build AI models. Ferreras said governments must outlaw non-disclosure and non-disparagement clauses that silence employees about safety risks and enact robust whistleblower protection. Tech workers also need collective standing over corporate strategy, independent channels for reporting safety breaches and formal representation inside frontier laboratories before models are deployed, she said.

On AI deployment, Ferreras cited employee backlash at Mark Zuckerberg’s firm, Meta, over the summer after Facebook, Instagram and WhatsApp introduced a tool that tracked staff keystrokes, mouse clicks and content displayed on computer screens to collect data for training the company’s AI models. Workers worried that privacy and consent were in ruins and that their jobs were on the line, she said. Her proposed response was a binding right to co-decision over workplace AI, under which no algorithmic system that monitors, evaluates or disciplines workers could be deployed without prior negotiation.

At national and EU level, Ferreras said policymakers need to consider collective bargaining on AI deployment and veto rights for works councils or trade unions, backed by independent technical experts and a strict ban on automated firing. She placed that proposal in the research of Oxford legal scholar Jeremias Adams-Prassl, who pioneered research into the gig economy and leads research into automated work in Europe. Adams-Prassl has shown that algorithmic management monitors, evaluates and dismisses workers across the socioeconomic spectrum, often without human oversight.

For corporate governance, Ferreras said legislators should require worker-elected directors and move toward dual-majority governance, under which strategic decisions require consent not only from capital but from labor. She also called for democratizing boardrooms and rewarding companies that refuse to concentrate power at the top. Ferreras argued that extending democracy into the firm is not an old battle over power for its own sake, but a practical necessity for more ethical judgment about the development and deployment of AI.

Ferreras argued that aligning AI with the good of workers and society requires more wisdom than closed executive rooms can offer. She cited 2024 Nobel laureates in economics Daron Acemoglu and Simon Johnson for the finding that technological revolutions do not automatically produce shared prosperity; institutions determine who benefits and who pays. She also invoked Yale political theorist Hélène Landemore’s concept of “democratic reason,” described as the collective intelligence of diverse, deliberating voices, and argued that this form of judgment should be brought to bear on corporate decision-making.

Ferreras concluded that leaving AI governance to self-regulation and the occasional whistleblower would entrench what she called an unaccountable algorithmic oligarchy. She argued that accountability and alignment begin with enforceable voice for both the people building AI systems and the workers whose work those systems govern, from the code floor to the boardroom. She also wrote that AI is “not merely a software product” but “a technology of government,” placing internal corporate power within the public debate over AI.