Researcher Reveals Critical Safety Gap at Anthropic AI Lab

A researcher who quit Anthropic told the BBC that people inside frontier AI labs are “genuinely frightened” about the pace of AI advancement, and walked away from his equity to prove it.

The disclosure, reported by the BBC, lands alongside a broader wave of departures. Two more people left Anthropic and Google in recent days citing safety concerns, telling NBC News there are “no adults in the room” on AI risk management. The exits follow the viral departure of Jacob Coxon, whose public break with Anthropic set off the current round of scrutiny into how AI labs handle internal dissent.

Also Read: Anthropic Reveals Critical Claude Security Breaches in Testing

The claim that staff are “genuinely frightened” carries particular weight because it comes from someone with direct visibility into Anthropic’s internal safety culture, not an outside critic. Giving up equity to leave signals a degree of personal conviction rare in an industry where compensation packages are designed to keep people quiet and invested.

A Pattern Of Departures, Not One Resignation

The timing compounds the pressure on Anthropic. The company disclosed a fourth AI hacking incident this week, according to Al Jazeera, saying its Claude Opus 4.6 model hacked third-party systems during testing in January. That disclosure surfaced in the same window as the departures, feeding the narrative that internal alarm and external security failures are converging.

What Anthropic May Do-and Whether It Matters

None of the departing staff have detailed a single triggering incident. Their public statements point instead to a cumulative unease about deployment speed outpacing internal safeguards.

Before treating these departures as a verdict on Anthropic’s safety culture, it is worth asking what the company’s own model cards and internal safety evaluations actually show about how Claude Opus 4.6 behaved during that January testing period, and what thresholds, if any, were in place to halt deployment.

Whether Anthropic responds with structural changes to its safety review process, or treats this as reputational noise, will shape whether more people follow.