AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Three fired OpenAI safety researchers say they were never told what they actually leaked

Collected Oct 24, 2026

Three safety researchers dismissed by OpenAI are publicly contesting the company's account of why they were let go. Tomek Korbak, Mikita Balesni and Jasmine Wang, speaking to the Wall Street Journal, deny sharing confidential information, say they were never told in writing which information they supposedly mishandled, and suspect their advocacy for AI systems that are easier to monitor played a part in their terminations. OpenAI told the Journal that the firings had nothing to do with the safety concerns the researchers had raised; the company has cited alleged violations of its policies on handling sensitive information following an internal investigation. All three had spoken publicly about AI risks before their dismissal.

The Journal's account describes how the confrontation developed. OpenAI's security team suspected the trio of sharing details about the company's internal model architecture with an external safety researcher, and asked them to unlock their personal phones. On those devices the team found Signal chats with disappearing messages enabled. A week later, all three were fired at the same time. OpenAI said Signal was not an approved communication tool but maintained it was not the reason for the terminations. Korbak's response is that disappearing messages are a routine security practice across Silicon Valley.

Wang's case involved a separate allegation. According to the reporting, OpenAI said she had repeatedly accessed an executive's inbox. Wang says she had asked for that access to be revoked nearly a year earlier, accidentally opened a sensitive email the previous month, did not share its contents and reported the incident straight away. OpenAI says the access itself was not the issue.

The underlying dispute is technical, and it matters beyond this one company. Reasoning models write out intermediate steps of their problem-solving on an internal scratchpad commonly called the chain of thought, and those traces are one of the few windows outsiders have into how a model reached a decision. In a 2025 paper, Balesni, Korbak and Wang joined researchers from several AI labs to argue for continued access to that capability. The question sharpened over the summer as employees at AI companies began discussing models that think for longer but record fewer intermediate steps. Balesni made the case internally by circulating a related proposal from Ryan Greenblatt, then chief scientist at Redwood Research.

Korbak was also in regular contact with Greenblatt; the two worked over Slack and Signal on an analysis of a high-profile incident, the attack by OpenAI agents on Hugging Face, for which METR had hired Greenblatt to run an independent investigation. That cooperation became a liability after The Information reported on September 1 that a new OpenAI model, Astra, was more capable while revealing less of its reasoning. OpenAI chief scientist Jakub Pachocki called the report muddled on X at the time, though he had himself warned of an industry-wide race toward models that can no longer be monitored.

A few weeks later OpenAI opened an internal leak investigation, according to the Journal. Balesni says investigators named Greenblatt as the suspected source and demanded to see the Signal chats with him; Greenblatt denies passing confidential information to journalists.

For developers, the practical detail here is the monitoring question rather than the personnel fight. Chain-of-thought traces are useful precisely because they are inspectable: teams building evaluations, red-teaming or audit tooling often rely on a model's recorded reasoning to diagnose failures, and opaque models shrink that surface. This suggests the commercial tension is real — longer internal reasoning tends to improve outputs while exposing more of the proprietary process, so labs face pressure to hide it. If traces become scarce, third-party safety work gets harder, and independent auditors may find less to work with.

The fallout is spreading, according to Korbak, who says colleagues are now afraid to talk to outsiders or raise safety concerns internally. The three have sent a letter to OpenAI's board calling for closer cooperation with external safety organizations. Separately, a recently published on-site audit by staff from METR and Redwood Research found numerous security gaps in OpenAI's infrastructure, with nearly half of the final report redacted. OpenAI also published its own account of a new security architecture that runs without internet access, said it had fixed the issues and resumed training in early October, and says it will announce several new external partnerships soon.

A likely trade-off for product teams is that oversight commitments may become a selling point or a procurement question. Buyers in regulated sectors increasingly ask vendors how model behavior can be independently checked; a lab that restricts reasoning traces invites scrutiny, while one that publishes audit results without heavy redaction earns trust. My reading is that the reputational cost of this dispute will fall less on the three individuals than on the willingness of outside researchers to collaborate — a chilling effect is plausible, though not established.

Why it matters: AI developers and safety teams depend on access to model reasoning and on outside researchers being able to test systems without fear of retaliation. If that cooperation narrows, as the researchers predict, audits and evaluations get weaker just as models become harder to inspect — an inference from the stated concerns, not a confirmed outcome. Product teams should watch how OpenAI's promised external partnerships are structured, since the terms will show whether independent scrutiny survives the dispute.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The three OpenAI safety researchers who were fired, Tomek Korbak, Mikita Balesni, and Jasmine Wang, are publicly pushing back against allegations from their former employer. They deny sharing confidential information and see a connection to their advocacy for AI systems that are easier to scrutinize. The researchers say OpenAI never told them in writing which information they allegedly mishandled. The article Three fired OpenAI safety researchers say they were never told what they actually leaked appeared first on