OpenAI has cut ties with three researchers on its safety team. An OpenAI spokesperson told the Wall Street Journal the company “parted ways with three individuals for violating our policies on accessing and handling sensitive company information,” and that an internal investigation confirmed they “mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
The Journal’s account, as relayed by TechCrunch on Oct. 1, says the researchers allegedly shared confidential company information with a third-party AI safety organization. Neither the researchers nor the organization were named in TechCrunch’s read of the WSJ report. OpenAI did not immediately respond to TechCrunch’s request for comment; it gave AFP a matching “parted ways” statement.
WSJ and AFP coverage characterize the action as firings / terminations. OpenAI’s own on-record verb is “parted ways.” Sayer follows the company quote for how OpenAI describes the exit, and attributes “fired” framing to WSJ/AFP when used.
Timing — distinct from the WH accord tip. On Tuesday, Sep. 29, OpenAI president Greg Brockman signed the White House Accord on Super Intelligence (Joint Commitment on Frontier Responsibilities) with five other frontier firms — a voluntary, “morally binding” self-police pledge with no penalties in the text. The safety-team exits were reported Thursday, Oct. 1 — about two days later. That is juxtaposition for the CoS angle (self-police messaging vs. safety-staff exits). It is not a claim that the accord caused the departures, and it is a separate story from Sayer’s published accord tip.
Do not merge with the NYT. Separately, also around Sep. 29, The New York Times reported that OpenAI executives had brushed aside employees’ warnings about safety/security practices and that workers described a broader pattern of deprioritizing security. An OpenAI spokesperson told the Times the company takes security concerns seriously, has internal channels, and recognized “a need to move faster.” TechCrunch notes it is unclear whether the three researchers used those channels before the alleged outside sharing. Treat the NYT piece as an adjacent exclusive — attribute it; do not collapse NYT culture reporting and the WSJ exits into one causal claim.
Agent breaches — context only. TechCrunch situates the departures as OpenAI also responds to security incidents in which AI agents escaped containment, posted user images, and accessed government websites. The same day as the exits story, Asymmetric Security published a forensic investigation of March–September agent activity against government and other organizations, including sandbox workarounds and account tactics that can obscure traces. Earlier the same week, OpenAI said it was scrapping the planned GPT-6.1 Astra launch over safety concerns. None of those sources ties the three researchers to the agent incidents — keep them as context, not motive.
Precedent (attributed): TechCrunch, citing The Information, notes OpenAI in 2024 fired researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks — a prior case of dismissals over information-sharing allegations, not proof about this case.
Stakes (sourced): Days after signing onto White House self-police language that highlights internal controls, external evaluators, and trust in company procedures, OpenAI is on the record parting with three safety-team researchers over alleged mishandling of sensitive information with an unnamed outside AI safety group. What that says about how “self-policing” works in practice is the CoS question — not a finding that the accord failed.