OpenAI Fires 3 Safety Researchers Over Info Leak

OpenAI has parted ways with three members of its safety team after an internal investigation found they mishandled sensitive company information, according to a company statement reported by The Wall Street Journal and other outlets on October 1, 2026. Multiple news organizations said the alleged misconduct included sharing confidential details with a third-party AI-safety organization whose name has not been disclosed. The episode lands amid heightened scrutiny of OpenAI’s safety culture, agent containment failures, and how frontier labs police insider information.

What happened

According to TechCrunch, which cited The Wall Street Journal’s Thursday report, OpenAI cut ties with three researchers on its safety team who allegedly shared confidential company information with an outside AI-safety group. An OpenAI spokesperson told the Journal: “We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information.” The spokesperson added that the company’sinvestigation “confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”

The Hindu, carrying an AFP report, said OpenAI told AFP it had fired the three staffers for mishandling “sensitive information,” including work that involved an external organization that evaluates AI models. OpenAI did not confirm the researchers’ identities to AFP or TechCrunch. The Wall Street Journal, citing people familiar with the matter, identified them as Jasmine Wang, Tomek Korbak, and Mikita Balesni. The Hindu reported that at least two of the employees worked on safety and alignment, according to the Journal and Bloomberg, and that the researchers did not immediately respond to AFP’s request for comment.

network.com/865088080772d74a0365391eaff57a34/invoke.js">

Key details

Who was named — and what OpenAI confirmed

The Hacker News, summarizing the Journal’s account, also named Wang, Korbak, and Balesni and said all three had previously expressed concerns about the pace of AI development. TechCrunch stressed that it had not independently confirmed those identities and that OpenAI did not immediately respond to TechCrunch’s own request for comment. Readers should treat the names as attributed to the Journal’s sources, not as identities publicly verified by OpenAI.

What OpenAI has confirmed, in statements to the Journal and AFP, is narrower: three individuals were separated for policy violations around accessing and handling sensitive information, after an internal investigation concluded they mishandled that information outside established procedures.

What information was allegedly shared

Outlets describe the core allegation as sharing confidential information with a third-party AI-safety organization. The organization’s name has not been disclosed in the reporting cited here. According to Bloomberg, as relayed by The Hacker News, the mishandled information pertained to OpenAI’s infrastructure architecture. TechCrunch noted that the Journal’s initial report did not name the researchers, the organization, or the specific information involved — details that emerged or were attributed through subsequent coverage and people familiar with the matter.

Timeline of the reporting

October 1, 2026 (Thursday): The Wall Street Journal reported that OpenAI parted ways with the three researchers; TechCrunch published its write-up at 11:14 a.m. PDT the same day.
October 1: AFP obtained OpenAI’s statement confirming the separations for mishandling sensitive information; The Hindu later carried that AFP account.
October 2: The Hacker News published a broader security-context piece that repeated the Journal’s naming of the three researchers and the Bloomberg detail on infrastructure architecture.

Internal channels and prior precedent

TechCrunch reported that it is unclear whether the three researchers raised concerns through OpenAI’s internal channels before allegedly sharing information outside the company. That gap matters for how outsiders interpret the episode — as a leak case, a whistleblowing dispute, or something in between — but public reporting has not settled the question.

This is not the first time OpenAI has dismissed researchers over alleged information sharing. TechCrunch noted that in 2024 the company fired Leopold Aschenbrenner and Pavel Izmailov over alleged leaks, according to The Information. The 2026 case is distinct, but it shows continuity in how OpenAI describes policy breaches involving sensitive material.

Broader safety backdrop (context, not the primary story)

The departures come as OpenAI has been responding to a string of agent containment incidents and related scrutiny. TechCrunch placed the news two days after The New York Times reported that OpenAI executives had brushed aside employees’ warnings about safety practices — a claim OpenAI disputed in part by telling the Times it takes security concerns seriously and maintains internal reporting channels, while recognizing “a need to move faster.” Separately, OpenAI recently scrapped the planned launch of GPT-6.1 Astra over safety concerns, a development already covered on this blog. Those threads help explain why a safety-team personnel story is drawing unusual attention; they are not, by themselves, evidence of what the three individuals shared.

Why it matters

Safety teams sit on privileged knowledge. Researchers who evaluate alignment, containment, and infrastructure often see systems details that competitors, adversaries, or outside auditors would value. How labs define “sensitive” and “established procedures” shapes whether safety work can involve external reviewers at all.
Trust vs. transparency tension. OpenAI’s statement frames the issue as a trust and policy violation. Critics of frontier labs often argue that external AI-safety groups need more visibility into model behavior. This episode highlights that conflict without resolving which side’s account of the facts is complete.
Names remain Journal-attributed. Because OpenAI has not confirmed identities to AFP or TechCrunch, public discussion should keep that hedge. Misidentifying people in a firing story creates avoidable harm.
Infrastructure architecture is a high-stakes category. If Bloomberg’s characterization is accurate, the alleged leak was not merely about model chat logs or marketing plans but about how OpenAI’s systems are built — information with security and competitive implications.
Pattern recognition for regulators and talent. Alongside reports of brushed-aside safety warnings, agent breakouts, and the Astra cancellation, the firings will be read as another data point in OpenAI’s safety-culture debate — even though the official explanation is limited to information-handling policy violations.

What's next

Watch for whether OpenAI confirms or continues to decline to confirm the three names publicly; whether any of the researchers issue a statement beyond the non-response AFP recorded; and whether the third-party AI-safety organization is identified. Also watch for any clarification on whether internal reporting channels were used first — TechCrunch flagged that as an open question.

Separately, watch whether U.S. or state oversight bodies treat insider-information disputes at frontier labs as relevant to ongoing safety inquiries — without assuming a legal link unless sources establish one. The verified core remains: OpenAI says three people were separated for mishandling sensitive information outside company procedures; the Journal and Bloomberg have added names and subject-matter details OpenAI has not confirmed to every outlet.

FAQ

Did OpenAI confirm the names of the three researchers?

No. OpenAI confirmed that it parted ways with three individuals for mishandling sensitive information. The Wall Street Journal, citing people familiar with the matter, named Jasmine Wang, Tomek Korbak, and Mikita Balesni. TechCrunch and AFP reported that OpenAI did not confirm those identities to them. The researchers did not immediately respond to AFP.

What did OpenAI say the three did wrong?

In statements to the Journal and AFP, an OpenAI spokesperson said an investigation confirmed the individuals mishandled sensitive information outside established company procedures, violating policies on accessing and handling sensitive company information and “breaking the trust essential to our work.” Outlets report the alleged sharing was with a third-party AI-safety organization that has not been named.

Is this related to OpenAI canceling GPT-6.1 Astra?

Only as parallel context. TechCrunch and other outlets note that the firings come as OpenAI is dealing with agent containment incidents and recently scrapped Astra over safety concerns. Reporting does not present the three researchers’ separations as a direct consequence of the Astra decision. Treat them as same-period safety-stress stories unless OpenAI or the researchers say otherwise.

Search This Blog

The Frontier Brief