AI Whistleblowers Tell New York: The Industry Is Moving Too Fast
AI Whistleblowers Tell New York: The Industry Is Moving Too Fast
The confrontation had been building for weeks. As scrutiny of frontier-model safety intensified, Council Speaker Julie Menin called a rare Committee of the Whole hearing, with all 51 members expected, to force major AI companies to explain the threats their systems could pose and discuss possible legislation. Anthropic, OpenAI, Google and Meta ultimately agreed to send senior policy and safety figures after the Council raised the prospect of subpoenas.
On Monday, the hearing brought together the industry’s defenders and some of its most prominent recent defectors. AI insiders urged city leaders to impose greater oversight, arguing that local governments could no longer stand aside as concerns mounted that advanced systems might endanger humanity.
Jacob Coxon, the former Anthropic employee whose resignation helped focus public attention on AI safety, framed the dispute as a clash between Silicon Valley’s standard operating model and technology with far higher stakes. AI firms, he said, carried a culture of “move fast, break things.” But, he added, “That works for a photo-sharing app. It does not work for the most powerful technology ever built.”
Coxon appeared alongside former Google and OpenAI employees, reinforcing the whistleblowers’ central argument: commercial competition is outpacing safeguards. Company representatives, meanwhile, faced pointed questions while seeking to defend their organisations’ approach to risk.
The sharpest exchange came when OpenAI policy executive Morgan Dwyer was asked to quantify AI’s danger. She said she did not know, while maintaining that no level of risk was acceptable to the company. That answer did little to calm councillors. Public Advocate Jumaane D. Williams called it a “God-awful, fantastically terrible answer.”
The hearing left the sides divided over the adequacy of corporate self-policing, but aligned on one point: the consequences of getting AI governance wrong could be severe. For the Council, that made the case for answers under oath—and potentially local rules—harder to dismiss.
Write a comment