Former OpenAI safety writer calls for stronger AI safeguards

David Robinson, who wrote safety reports for major OpenAI model releases, resigned this week and has used an essay in The Atlantic to criticise the culture of the AI industry.

Robinson argues that “extreme confidence” and “perpetual sprints” to build larger and more capable models can leave companies overlooking or underestimating potential problems. He describes the industry’s approach as “unimpeded optimism”.

He says the issue runs deeper than adding rules or regulations for model training. In his view, AI companies need greater humility and should look beyond the technology sector’s “move-fast-and-break-things” culture.

Robinson argues that frontier AI labs should operate with safeguards comparable to those at nuclear power plants or busy airports. He calls for layers of redundancy and careful planning to reduce the chance that human error leads to disaster.

The Verge also reports departures by staff at Anthropic and Google DeepMind, but provides no detail establishing whether those departures shared Robinson’s reasons.