When tech workers have to quit their jobs to raise the alarm, it’s time to worry Source: theguardian.com.
Over the past two years, a troubling pattern has emerged in leading AI labs worldwide. Safety researchers and engineers have repeatedly resigned to issue public warnings about unchecked AI development.
Last month, a senior safety officer at OpenAI stepped down. He called the company’s culture ‘broken,’ saying safety was being sacrificed for speed and profit. He warned the entire AI industry lacks proper safeguards, risking catastrophic outcomes.
The previous month, a researcher at Anthropic—OpenAI’s rival—quit after warning that advanced AI could ‘kill us all by the end of the decade’ if not properly governed.
Exit Is Not Voice
These resignations aren’t personal choices. They reveal a systemic failure in how AI companies are governed.
As political scientist Albert Hirschman wrote, ‘exit is not voice.’ When people leave to speak up, it shows the system lacks internal accountability.
Each departure takes with it specialized knowledge and institutional memory. Without these voices, companies lose the ability to foresee and prevent real-world risks.
AI is not just software. It is a technology of government—shaping public services, policies, and information access.
In labs like OpenAI, Anthropic, Microsoft, and Google, a small group of executives and investors decide which values go into AI models. These values affect how AI interacts with millions of users and influences society.
Yet corporate governance still treats firms as owned by shareholders. Workers are seen as subordinates, not stakeholders.
Workers contribute their time, intellect, and health. But they rarely shape strategic decisions. Research by Oxford legal scholar Jeremias Adams-Prassl shows algorithmic systems already monitor, evaluate, and discipline employees—often without human oversight or consent.
In a democratic society, governance must reflect the consent of the governed. Current corporate structures fail this principle.
Three Key Reforms for Democratic AI
First, governments must outlaw non-disclosure and non-disparagement clauses. These silence employees about safety risks.
Workers need legal protection to report concerns. This includes strong whistleblower protections and independent reporting channels outside corporate control.
Second, AI deployment must include co-decision rights for workers. No monitoring or evaluation system should be used without prior negotiation.
Workers must have a say in AI tools that track keystrokes or screen activity. At national and EU levels, collective bargaining and veto rights for works councils or trade unions should be established.
These rights must be backed by independent experts and a strict ban on automated firing.
Third, we must democratize boardrooms. Firms should include worker-elected directors on their boards.
A dual-majority governance model—requiring both capital and labor consent—ensures decisions reflect broader societal values.
As Nobel laureates Daron Acemoglu and Simon Johnson have shown, technology does not automatically bring shared prosperity. Institutions determine who benefits and who bears the cost.
Without democratic participation, AI development risks becoming the domain of a small elite—an unaccountable algorithmic oligarchy.
Accountability begins not with external rules alone, but with giving those who build AI—and those whose work it governs—an enforceable voice, from the code floor to the boardroom.
Isabelle Ferreras, an accelerator fellow at the University of Oxford’s Institute for Ethics in AI and co-founder of DemocraticIntelligence.org, writes this as a call for structural change. She argues public regulation alone is insufficient. Real progress requires democratic principles embedded in corporate governance.
For more on how power is reshaping technology, see Rightwing ‘Great Replacement’ Fears Mask Tech’s Real Great Replacement.
For insights on training and employment, explore UK to Raise Training Costs for Older Workers to Boost Youth Apprenticeships.
When workers are silenced, the system fails. The real test of AI safety is not just in its code—but in who governs it and how.
Sources & further reading
Featured image: original publisher image. Via original publisher. Image source
