OpenAI Safety Staff Member Resigns, Raises Concerns

David Robinson, who previously authored the safety reports for major model releases at OpenAI, has stepped down from his role and is now sharing his thoughts in an editorial.

It’s easy to feel skeptical about the sudden influx of voices warning about the dangers of the technology they helped create. After all, they played a part in shaping this landscape. However, their concerns shouldn’t be dismissed.

Robinson argues that the industry’s culture is fundamentally flawed, suggesting that the issues at hand go beyond simply implementing new rules or regulations for training models. He points out that Silicon Valley has thrived on “extreme confidence” and “perpetual sprints,” focusing on creating larger and more advanced models while overlooking potential risks.

He emphasizes the urgent need for AI companies to adopt a sense of humility and step outside the fast-paced, insular tech environment. Robinson advocates for safeguards akin to those in nuclear power plants, stating:
> Given today’s risks, frontier labs need to run like nuclear power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster.

Robinson is part of a growing trend of researchers and safety professionals leaving their roles at leading AI companies. Jacob Coxon initiated this wave by resigning from Anthropic and publicly warning that AI “could kill us all by the end of the decade.” He was soon joined by Robert O’Callahan, Bilal Chughtai, and Josh Engels from Google DeepMind, along with Joe Benton from Anthropic.