# Former OpenAI safety researchers ask oversight bodies to protect outside collaboration > Three dismissed OpenAI researchers urge oversight bodies to protect outside safety work and clarify employee rules. OpenAI disputes their account of the firings. By BIG CHANGE Editorial Published: 2026-10-09T04:59:01.631Z Updated: 2026-10-09T04:59:01.631Z Canonical: https://bigchange.ai/blog/openai-researchers-letter-outside-safety-evaluators ![A blank ivory letter sheet lies open over three aligned pages on a warm ivory background.](https://bigchange.ai/api/media/file/openai-monitorability-letter-hero-v2.png) Conceptual illustration of the researchers’ public open letter calling for clear rules for outside safety work. It does not reproduce the letter or depict internal events or a model assessment. AI-generated illustration by BIG CHANGE. Three former OpenAI safety researchers have asked the company’s oversight bodies to preserve independent safety work and clarify how employees may share information with outside evaluators. In an [open letter published October 8](https://mikitabalesni.com/letter/letter.pdf), Tomek Korbak, Jasmine Wang and Mikita Balesni dispute the account of misconduct that followed their dismissal. OpenAI told TechCrunch that an investigation found a pattern of mishandling research information beyond sharing with an outside evaluation group. The letter and TechCrunch report do not provide the underlying investigation record or the specific policies at issue. The letter is advocacy by the three people affected. It says they believed their work with external groups fell within their jobs and OpenAI’s working norms. They deny being the source of a leak to The Information about supposedly less monitorable model architectures. Those statements describe their account of events, not findings established by an independent inquiry. ## The big change - **What changed:** The researchers have asked OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council to address how employees can work with independent safety evaluators while respecting confidentiality. - **Why it matters:** Their letter asks for continued auditor access, protection of model monitorability and explicit rules for outside collaboration. These requests concern both the information outside evaluators can receive and the boundaries employees must follow when sharing it. - **What to watch:** OpenAI’s reported response says the firings followed policy violations and were unrelated to raising safety concerns. Its response, as reported by TechCrunch, does not identify the specific policies or evidence behind the decisions. ## What the researchers say happened Korbak says he was a point of contact for the external evaluator METR during an investigation of an OpenAI agent incident involving Hugging Face. The letter says the investigation had no established precedent and its internal procedures were developing at the time. It says he tried to work within the policies then in force while communicating with outside counterparts. Balesni says he worked on cross-company commitments concerning the ability to monitor models’ reasoning. According to the letter, he discussed that work with board members and executives, checked with his reporting line and removed sensitive details before sharing material. Wang says access to an executive’s email had been delegated for recruiting; after she accidentally opened a sensitive email, she says, she promptly reported it and again asked IT to remove the access. None of those accounts establishes what OpenAI’s investigation found. ## The requests and the company’s response The letter makes three recommendations. First, OpenAI should follow through on a public commitment, which the researchers attribute to Sam Altman, to give independent evaluators ongoing access. They express concern that the firings could narrow work with METR or other auditors. Second, they ask OpenAI and other frontier model developers to avoid changes that further reduce monitorability while safety work relies on it. The letter refers specifically to monitoring chain-of-thought reasoning; it does not establish a measured decline in a named model. Third, they ask OpenAI to reaffirm open safety discussion and publish clear procedures for working with external organizations. OpenAI has not issued a formal response to the letter, according to [TechCrunch’s October 8 report](https://techcrunch.com/2026/10/08/fired-openai-safety-researchers-dispute-misconduct-claims-warn-of-chilling-effect/). The publication says an internal memo shared with it denied that the dismissals were retaliation for speaking out and agreed with the researchers’ recommendations. Separately, a spokesperson told TechCrunch that an investigation found a pattern of misconduct in handling research information that went beyond outside sharing. TechCrunch says OpenAI did not answer its questions about the particular policies, the circumstances of the dismissals or protections for employees who work with external evaluators. The public record shows agreement in principle on the value of outside safety work, alongside a dispute over how these three employees conducted it. The letter cannot verify its authors’ permissions or intentions, and the company’s reported statements do not let readers assess the alleged violations. OpenAI’s oversight bodies still have to address what rules govern such collaboration and what access independent auditors will have. ## Sources & further reading - [Korbak, Wang and Balesni, “OpenAI cannot make AI safe on its own”](https://mikitabalesni.com/letter/letter.pdf), open letter published October 8, 2026. Primary source for the authors’ account and their three recommendations. Their descriptions of internal events, permissions and model work are claims by interested parties. - [Rebecca Bellan, TechCrunch](https://techcrunch.com/2026/10/08/fired-openai-safety-researchers-dispute-misconduct-claims-warn-of-chilling-effect/), October 8, 2026. Reports OpenAI’s spokesperson and internal memo responses and the questions the company did not answer. It corroborates what each side said, not the underlying disputed conduct. ## Sources - [OpenAI cannot make AI safe on its own](https://mikitabalesni.com/letter/letter.pdf) — Primary letter; the signers’ disputed internal-event claims are attributed. - [Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect](https://techcrunch.com/2026/10/08/fired-openai-safety-researchers-dispute-misconduct-claims-warn-of-chilling-effect/) — Rebecca Bellan’s reporting on the letter and OpenAI’s spokesperson and memo response.