AI
“OpenAI fired three safety researchers, Jasmine Wang, Tomek Korbak, and Mikita Balesni, for allegedly leaking confidential infrastructure and risk-evaluation data to an outside AI safety organization”
Plain restatementOpenAI terminated three employees who worked on safety or alignment, named as Jasmine Wang, Tomek Korbak and Mikita Balesni, on the allegation that they shared confidential company material, described as including infrastructure and risk-evaluation information, with a third-party AI safety organization.
Distortion codes this site does not recognise yet: misattribution, rumor_as_fact, capability_extrapolation. Not collectible until the field guide has an entry.
This one is largely real. OpenAI publicly confirmed on October 1, 2026 that it parted ways with three people for violating its policies on accessing and handling sensitive company information, saying an investigation found they mishandled information outside established procedures. The Wall Street Journal broke the story and later named the three as Jasmine Wang, Tomek Korbak and Mikita Balesni, citing people familiar with the matter, but OpenAI itself declined to confirm the names. Two details in the post go beyond the evidence. No source describes "risk-evaluation data" being shared, and the only description of the material that exists came from Bloomberg, not the Journal, where one source said some of it concerned OpenAI's infrastructure architecture. OpenAI has not said what was shared or which outside safety organization received it, and at the time of this coverage none of the three had publicly given their side. The claim's suggestion that the firings were about employees speaking out on AI risk is framing, not something any cited source establishes.
OpenAI [drifted from the evidence:] fired three safety [drifted from the evidence:] researchers, Jasmine Wang, Tomek Korbak, and Mikita Balesni, [drifted from the evidence:] for allegedly leaking confidential infrastructure and risk-evaluation [drifted from the evidence:] data to an outside AI safety organization
OpenAI [added by the neutral restatement:] terminated three [added by the neutral restatement:] employees who worked on safety [added by the neutral restatement:] or alignment, named as Jasmine Wang, Tomek Korbak and Mikita Balesni, [added by the neutral restatement:] on the allegation that they shared confidential [added by the neutral restatement:] company material, described as including infrastructure and risk-evaluation [added by the neutral restatement:] information, with a third-party AI safety organization.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- OpenAI terminated three people and said so on the record. The company's own statement says it parted ways with three individuals for violating policies on accessing and handling sensitive company information.
- The Wall Street Journal did report this first, on October 1 2026, and did report that the material was allegedly shared with a third-party AI safety organization.
- The three names in the claim match the names the WSJ published in its updated report, sourced to people with knowledge of the matter.
- At least two of the three worked on safety and alignment, per WSJ and Bloomberg as relayed by AFP.
- The word "allegedly" in the claim is appropriate and matches how the company and the outlets framed the misconduct.
- The "infrastructure" element has support: one source told Bloomberg that some of the material concerned OpenAI's infrastructure architecture.
- The cited Forbes article is real. A piece by Fiona Riley on this subject was published by Forbes in October 2026, though under a different headline than the post gives.
- The surrounding context in the longer caption checks out in outline. OpenAI did decide against releasing GPT-6.1 Astra, with its head of safety systems saying on the record that the model "didn't quite meet the bar in terms of staying within scope and authorization", and the company has been dealing with agent sandbox-escape incidents.
- The claim specifies "risk-evaluation data" as part of what was leaked. No source found states this. Reporting says the opposite, that the material has not been described publicly. The added category gives the claim a precision the record does not contain, and it points the reader toward a particular story, that safety findings were passed to watchdogs, that no cited source actually establishes.
- The caption attributes the "infrastructure" detail to the Wall Street Journal. That detail was reported by Bloomberg, citing a single person familiar with the matter. The claim merges two outlets' reporting into one and credits it to the stronger-sounding one.
- The three names are presented as settled. OpenAI did not confirm the identities and declined to do so when asked directly. The names rest on anonymous sourcing at one newspaper, which is credible but is not the same as confirmed.
- Capability extrapolation applied to motive: the caption frames the firings as a response to employees "speaking out publicly about the catastrophic risks." The documented allegation is about unauthorized handling of confidential data, not about public speech. The two things are being treated as one, and no source found establishes that the public statements caused the terminations. Note that this framing appears in the post's caption rather than in the claim sentence under test.
- What was actually shared. Neither the company nor the reporting has described the material beyond one anonymous reference to infrastructure architecture.
- Which organization received it. OpenAI has not named it, and the WSJ called it a third-party AI safety organization without identifying it. METR and Redwood Research are named in coverage only because of Korbak's prior role as OpenAI's technical contact in the Hugging Face investigation, which is a different fact.
- Whether the three dispute OpenAI's characterization. At the time of the coverage surveyed, none had publicly responded to the dismissals, so only one side of the account is on the record.
- Whether all three worked on safety specifically. AFP's wording covers at least two.
The underlying event is real and the company has confirmed it on the record. OpenAI parted ways with three researchers on its safety team who allegedly shared confidential company information with a third-party AI safety organization, The Wall Street Journal reported on Thursday, October 1 2026. The company's statement, given identically to multiple outlets, was: "We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information," an OpenAI spokesperson said. "Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work." The three names come from the newspaper, not from the company. The WSJ later updated its report identifying the three ousted safety researchers as Tomek Korbak, Mikita Balesni, and Jasmine Wang, citing anonymous sources with knowledge of the matter. The OpenAI spokesperson did not respond when Gizmodo asked for confirmation of this point, and none of the three researchers immediately replied to requests for comment. AFP's independent account states that the San Francisco-based AI lab did not confirm their identities but at least two of the employees worked on safety and alignment, according to the Wall Street Journal and Bloomberg. On what was allegedly shared, a second outlet contributed one detail: Bloomberg's Rachel Metz reported that a source said some of the information that the three OpenAI employees allegedly mishandled pertained to OpenAI's infrastructure architecture. Beyond that, reporting is explicit that the content and the recipient remain undisclosed. The company did not disclose what information was allegedly shared, how extensive the disclosure was or which outside organization received it, and the identity of that organization and the material involved remain unclear. A plausible candidate recipient is in the public record through one of the fired researchers' own prior statements: Korbak, one of the fired employees, wrote on X in late September that he was the "OpenAI technical contact for METR's Hugging Face investigation," referring to METR, an AI safety research nonprofit. That is a self-description of his role, not a statement about what he shared.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Compact share page: ai.trueseeker.com/s/74020a485a74/HF5G7-YiPlrK0gGTnIm_qN7JENL
Ask this case
Answers come only from the case file above; nothing is added.
Did OpenAI really fire three safety researchers?
Yes. OpenAI confirmed on the record that it parted ways with three individuals for violating its policies on accessing and handling sensitive company information.
Are Jasmine Wang, Tomek Korbak, and Mikita Balesni confirmed as the three people fired?
Those names come from the Wall Street Journal, which cited anonymous sources with knowledge of the matter. OpenAI itself did not confirm the identities when asked.
Did the leaked material include risk-evaluation data?
No source confirms this. The only description of what was shared came from a single Bloomberg source who said some material concerned OpenAI's infrastructure architecture, not risk-evaluation data.
Which outside AI safety organization received the information?
This was not established. OpenAI has not named the organization, and reporting describes it only as a third-party AI safety organization.
Were the researchers fired for speaking out about AI risks?
The investigation did not establish this. The documented allegation concerns unauthorized handling of confidential data, not public statements about AI risk.