OpenAI alleges further undisclosed breaches by fired safety researchers

OpenAI stands by the dismissals and says contracts with outside safety assessors are being finalised. The researchers allege retaliation, which the company denies.

OpenAI stands by the dismissals and says contracts with outside safety assessors are being finalised. The researchers allege retaliation, which the company denies.
On 9 October, OpenAI defended the dismissals, saying its investigation found breaches beyond those described in the researchers' open letter, but did not disclose details. Jasmine Wang, Tomek Korbak and Mikita Balesni, three safety researchers fired by ChatGPT-maker OpenAI, deny misconduct and say they believe they were dismissed for raising safety concerns.
TechCrunch first reported the dismissals on 1 October, citing The Wall Street Journal's account that three safety-team researchers allegedly shared confidential information with an outside superintelligence safety organization. On 2 October, The Hacker News relayed the Journal's identification of the researchers, citing people familiar with the matter, and Bloomberg's report that the information concerned OpenAI's infrastructure architecture—the structure of its technical systems. OpenAI told the BBC its investigation confirmed information mishandling outside established procedures. The BBC reported on 2 October that at least two employees worked in safety research and that it understood they were dismissed for alleged information mishandling, not for raising safety concerns.
In its 9 October response, OpenAI said the researchers violated clear policies on handling sensitive information. A note from research leaders said the internal investigation uncovered a “significant breach of trust” beyond what the researchers described in their letter, published on 8 October. OpenAI stood by its decision but did not disclose details of those additional findings.
The company again denied that the dismissals were about raising safety concerns or speaking out. An internal memo OpenAI shared with TechCrunch for its 8 October report had also denied retaliation, praised the researchers' safety contributions and said the company encourages employees to raise concerns. TechCrunch reported that OpenAI had not directly answered its questions about the specific policies allegedly violated, the dismissal circumstances or protections for employees who raise concerns and collaborate with outside evaluators.
The researchers' letter denies contact with outside parties beyond their job mandates. It says Korbak believed his communication with outside safety evaluators during a sensitive investigation followed company policies and norms, while policies were being developed as the investigation proceeded. Balesni worked on model monitorability—how readily researchers can examine a model's reasoning. The letter says that work required extensive outside communication, had support and coordination from board members and executives, and involved checks with his managers and removal of sensitive details before sharing materials. The three also denied involvement in a leak to The Information about designs in OpenAI's newest models that make their reasoning harder to monitor.
Wang separately said OpenAI told her she was dismissed for accessing an executive's email, access she said the company had delegated for recruiting. She said the information technology team did not act on her request to remove it when she no longer needed it, and she could not remove it herself. The executive's inbox appeared indistinguishably alongside her own in her phone's mail app, she said. After accidentally opening a sensitive email, she said she notified the executive within minutes and again requested removal of the access.
The researchers say the abrupt dismissals have made former colleagues afraid to speak or collaborate in ways previously accepted at OpenAI. Their letter warns that fear and unclear rules could impede safety research and outside scrutiny.
The dispute follows criticism of OpenAI's handling of safety warnings. According to TechCrunch, employees told The New York Times in a report published on 29 September that executives had brushed aside warnings and described a broader pattern of deprioritizing security. OpenAI told the Times it takes security concerns seriously, provides internal reporting channels and recognized “a need to move faster.”
The outside recipient organization remains unnamed, and the exact information allegedly shared and detailed investigation findings remain undisclosed. It is still unclear whether the information concerned safety issues or whether the researchers used internal reporting channels before the alleged outside sharing.
The researchers called on OpenAI to uphold its public commitments to embed outside safety auditors within the organization, preserve the ability to monitor its most advanced models and support open dialogue with outside safety researchers. The internal memo shared with TechCrunch said OpenAI agreed with those recommendations.
In its 9 October response, OpenAI said it was finalising contracts with third-party safety assessors—outside specialists who evaluate safety—and would announce details in the coming weeks. The company described contracts still being finalised, not completed appointments.

IMPEL and IMPF want songs generally valued at least as highly as recordings, with payments and disclosure covering past use, SI training and generated music.
3 min

City police claimed five cameras, while the owners of seven cameras remain publicly unidentified. The county says it will not investigate ownership, citing limited resources.

Developers can collect streaming data, manage it in a catalog and query it with Cloudflare or outside tools. Free transfers out of R2 do not eliminate connected-service or pipeline delivery charges.
2 min