Fired OpenAI researchers dispute "pattern of misconduct"

The dispute arrives on top of a run of earlier internal friction at OpenAI. In May 2024, WIRED reported that OpenAI's Superalignment team had disbanded, with its work absorbed into other research efforts. Then in an Oct. 3 essay, David Robinson said he had resigned that week after leading the writing of OpenAI's launch safety reports, describing its culture as "broken." Read together, these episodes help explain why the fired researchers cast their dismissals as a signal about how the company treats safety work.

OpenAI had already set out principles for outside assessments on Sept. 22, calling for deep access for assessors and independence over their conclusions and reports, alongside agreed limits, security and intellectual-property restrictions, and procedures for redacting material. It said it was discussing assessments with several third parties but named no embedded partner. The researchers' worry is about a possible future restriction — that the firings could give OpenAI cover to pull back from embedding outside auditors.

The contrast with Anthropic is a useful yardstick. Anthropic named a partner, Accenture, for embedded evaluators with access comparable to employees, including visibility into training, development decisions and staff discussions, and said it would fund Accenture's work directly while acknowledging unsettled standards for access and reporting. For OpenAI, the practical next step the researchers' concerns point to would be naming assessors, stating when their work starts and explaining what they can inspect — details that would let staff and outside observers judge how far that scrutiny reaches.