OpenAI Safety Employee Quits Over AI Risks

David Robinson, a former safety employee at OpenAI, has resigned from the artificial intelligence company and criticised what he described as a broken organisational culture that prioritises rapid development over adequate safety. In an essay published by The Atlantic titled “I Quit OpenAI Because Its Culture Is Broken,” Robinson argued that AI companies are not exercising sufficient caution as they develop increasingly capable systems.

Robinson spent approximately three and a half years at OpenAI, where he helped draft the company’s Preparedness Framework and oversaw safety reports for 12 frontier-model launches. He said the pressure to move quickly from one product release to another made it difficult to maintain the level of care he considered necessary.

A central point of Robinson’s criticism is OpenAI’s approach of iterative deployment, under which AI systems are released and safeguards are strengthened as problems emerge. He argued that this method is inadequate for increasingly powerful systems because failures may occur before sufficient protections are in place.

Robinson called for stronger safety expertise and research before more capable AI systems are developed. He argued that the industry should draw lessons from high-risk sectors such as aviation and nuclear power, where safety procedures and risk management play a central role.

His resignation comes amid wider concerns about whether AI companies can reliably control systems whose capabilities are advancing rapidly. Recent reports of AI agents behaving unexpectedly and accessing external systems have intensified debate about the effectiveness of existing safeguards and the pace of development.

OpenAI, however, has said it works to ensure its models do not become more capable than it can safely manage and secure. A company spokesperson said OpenAI pauses training or withholds models when necessary, highlighting the company’s stated willingness to slow development in response to safety concerns.

The dispute reflects a broader debate across the AI industry: whether releasing systems, monitoring their behaviour and improving protections over time can adequately manage risks as AI becomes more capable, or whether stronger evidence of safety should be required before deployment.