TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Three OpenAI safety researchers say their recent firings followed an investigation tied to the Hugging Face incident and warn the dismissals could discourage staff from raising concerns. OpenAI says an internal investigation found violations of policies on sensitive information, denies firing anyone for raising safety concerns, and has not publicly detailed the alleged breach.
Three OpenAI safety researchers say the company’s recent decision to fire them has intensified concerns about whether employees can raise safety issues without risking their jobs. OpenAI says an internal investigation found violations of its policies for handling sensitive information, but the company has not publicly described the specific conduct it says justified the dismissals.
In an open letter addressed to OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Council, the researchers said the firings were abrupt and had left colleagues unsure which forms of work or communication might lead to dismissal. “If conduct that was considered normal last month now constitutes grounds for sudden dismissal, everyone at OpenAI is left guessing where the line is,” the letter said.
Tomek Korbak, one of the researchers, wrote on X that he was told by the head of OpenAI’s safety department that the company no longer trusted him. He said he was escorted out after a security officer took his badge. Korbak said he was verbally told the decision concerned how he communicated with METR, an external safety lab involved in examining the Hugging Face incident. He said no specific explanation was given in writing. “To be clear, talking to METR was my job,” he wrote.
The source report says Korbak and Mikita Balesni worked directly on the incident investigation; Balesni also worked on industry commitments related to monitoring AI models. The third researcher, Jasmine Wang, said the group’s account connected her firing to an incident in which she accidentally accessed a sensitive email through delegated recruiting access and reported it within minutes. Those descriptions come from the researchers’ account and have not been independently established in the material provided.
The Dispute Over Safety Oversight
The disagreement is about more than three employment decisions. It puts attention on how OpenAI handles internal safety concerns, how staff can cooperate with outside auditors, and whether monitoring tools will remain available for advanced models. The researchers argue that fear of dismissal could discourage employees from reporting risks or sharing information with external safety specialists.
They also argue that monitoring a model’s internal reasoning, often called chain-of-thought monitorability, is one tool for detecting unsafe behavior. They say the industry does not yet know how to safely build models that cannot be monitored. That is their assessment, not a finding independently verified by the source material. If monitoring becomes harder while external scrutiny is constrained, safety teams and outside reviewers could have less ability to identify problems before systems are deployed.
OpenAI’s response leaves readers with two competing accounts: the researchers describe unclear rules and retaliation fears, while the company says the firings followed a serious breach of trust. The outcome matters to employees, external safety groups and users who depend on credible oversight of increasingly capable AI systems.
As an affiliate, we earn on qualifying purchases.
From Safety Criticism to Firings
The dispute follows earlier criticism of OpenAI’s safety culture. In May 2024, Jan Leike, then a leader in the company’s superalignment safety work, left for Anthropic and publicly argued that safety processes were falling behind product development. That episode is relevant background, but it does not establish the reasons for the current firings.
The latest controversy centers on a Hugging Face security incident and the internal investigation that followed. The source report says the researchers described the investigation as unusual, with procedures being developed while it was underway. They deny being the source of a report about allegedly less-monitorable AI architectures, saying that coverage undermined work on industry-wide restrictions. They also say they were not asked to respond to an alleged board-level memo related to the matter.
The researchers’ letter calls on OpenAI to honor commitments to give external safety auditors, including METR, access inside the organization; preserve monitorability in frontier models; and spell out rules for employee collaboration with outside safety groups. OpenAI has said it is working on contracts with external safety auditors and agrees that model monitorability requires an industry-wide commitment.
“To be clear, talking to METR was my job.”
— Tomek Korbak, in a post on X
As an affiliate, we earn on qualifying purchases.
What the Investigation Has Not Shown
OpenAI has not publicly explained what sensitive information was mishandled, what additional breach it says investigators found, or which policies the researchers allegedly violated. The researchers, for their part, say they followed the norms in place during the investigation and deny leaking information about less-monitorable architectures. The source material does not independently establish either account.
It is also unclear whether the company gave the researchers a written explanation, what evidence the internal investigation relied on, or whether they had a formal chance to respond. The source report provides no detailed timeline for the firings or the investigation, and OpenAI’s stated denial does not resolve the researchers’ concerns about how safety work with external groups should be handled.
sensitive information security devices
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Auditor Rules and Monitoring Commitments
OpenAI says it is working on contracts with external safety auditors and supports an industry-wide commitment to model monitorability. The researchers are asking the company to turn those positions into practical commitments: auditor access, protections for monitoring frontier models, and clear written guidance for staff working with outside safety groups.
The next developments to watch are whether OpenAI publishes more detail about the investigation, clarifies its policies, or announces specific auditor arrangements. The source material does not give a timetable for those steps. Until the company provides further information, the dispute over the firings and the effect they may have on employees’ willingness to report safety concerns remains unresolved.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why were the three OpenAI researchers fired?
OpenAI says an internal investigation found violations of policies for handling sensitive information and a significant breach of trust. The company has not publicly specified the conduct. The researchers dispute the account and say the reasons given to them were unclear.
Did OpenAI say the firings were retaliation for safety concerns?
No. OpenAI said it does not terminate employees for raising concerns. The researchers argue that the circumstances could make other staff afraid to speak up, but the material provided does not establish that safety advocacy caused the dismissals.
What was the Hugging Face incident?
The source report describes it as a security incident investigated by OpenAI, with Korbak and Balesni involved in the inquiry. It does not provide enough detail to independently characterize the incident or its full scope.
What changes are the researchers asking OpenAI to make?
They want external safety auditors such as METR to receive access inside the organization, frontier models to remain monitorable, and clear rules for employees working with outside safety groups.
What is still unknown?
OpenAI has not detailed the alleged policy violations or the additional breach of trust it cited. The available material also does not establish whether the researchers’ accounts or the company’s findings will be independently reviewed.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
