An OpenAI safety employee has quit and is sounding the alarm
David Robinson, who authored OpenAI's safety reports for major model releases, has resigned and published a warning in The Atlantic about AI development culture. His departure is part of a growing exodus of safety researchers from leading AI companies, highlighting growing concerns about the pace and direction of AI advancement.
Robinson argues that Silicon Valley's "extreme confidence" and move-fast culture is fundamentally broken, with companies building larger AI models with "unimpeded optimism" while ignoring potential risks. He is calling for AI firms to adopt "nuclear-level safeguards" akin to nuclear power plants and airports, with layers of redundancy and careful planning. Robinson's departure follows similar resignations by Jacob Coxon from Anthropic, Robert O'Callahan and Bilal Chughtai from Google DeepMind, and Joe Benton from Anthropic, all raising public concerns about AI safety.
- OpenAI safety researcher David Robinson quits, warns of dangerous industry culture.
- He's calling for nuclear-level safeguards and a slower, more cautious approach to AI.
- Part of a wave of safety workers leaving major AI companies over concerns.
New here? Start with this
Artificial intelligence is advancing rapidly through companies like OpenAI, which develops increasingly powerful systems used globally. As these systems become more capable, questions have emerged about whether companies are building and testing them carefully enough, and whether potential harms are being adequately addressed.
There is disagreement within the industry about the appropriate balance between rapid development and caution. Some safety researchers contend that major AI companies prioritise speed and capability over careful evaluation of risks, developing larger models whilst insufficient safeguards are in place.
This matters because researchers explicitly hired to evaluate safety and potential problems have recently started leaving these companies and raising public concerns. Their departures suggest tensions between safety-focused staff and company priorities, raising questions about how seriously risks are being considered in the AI industry's most influential organisations.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The resignation of experienced safety researchers points to a genuine structural problem in how AI companies approach risk. These are experts with insider knowledge, and their exodus signals that current safeguards are inadequate for the scale and speed of development. The move-fast-and-break-things culture, whilst valuable for consumer software, is inappropriate for technology that could pose existential risks. Implementing rigorous, layered safety frameworks like those in nuclear power or aviation is not merely prudent but essential before deploying increasingly capable systems.
The case against
Whilst safety concerns warrant serious attention, the current approach reflects a reasonable balance between innovation and responsible development. AI companies employ substantial safety teams, conduct red-teaming exercises, and publish research openly. Demanding nuclear-level safeguards might impede progress on beneficial applications—healthcare, scientific research, education—and could disadvantage careful companies relative to less scrupulous international competitors. The market itself incentivises safety; companies facing public criticism and regulatory scrutiny have strong reasons to address genuine risks without needing the extreme caution Robinson advocates.