OpenAI fires three safety researchers, disputes their account

By Splitfeed EditorialLast updated Oct 9, 2026, 10:00 UTC

On or around October 7, 2026, OpenAI dismissed safety researchers Jasmine Wang, Tomek Korbak, and Mikita Balesni, citing what it called a significant breach of trust involving mishandling of sensitive information. OpenAI posted a statement on X explicitly saying the firings were not connected to safety advocacy or speaking out. Around the same time, a separate senior OpenAI safety staffer resigned and spoke publicly about their reasons. The three fired researchers responded with an open letter to OpenAI's board denying they leaked anything, asserting they were dismissed for prioritizing safety over the company's near-term interests, and calling on the board to preserve chain-of-thought monitoring, pursue independent safety audits, and take seriously the risk that OpenAI may lose the ability to control its AI agents.

The dispute matters because chain-of-thought monitoring is one of the few tools that gives researchers visibility into how AI models reason through problems. If it is reduced or eliminated, critics argue that AI systems become harder to audit and correct. The firings also arrive at a moment when OpenAI is under sustained scrutiny over whether its safety commitments keep pace with its commercial ambitions, making the question of who gets to raise internal concerns, and how, genuinely consequential.

The anti-AI camp is treating the dismissals as confirmation that OpenAI punishes safety prioritization. Mikita Balesni stated publicly that he believes the three were fired for prioritizing safety over the company's near-term interests, and accounts including Engadget and Paul Lander highlighted the researchers' own warning that the firings will have a chilling effect on remaining employees. The pro-AI camp has not yet published reactions; a camp holding that position would typically argue that companies need to enforce information-handling rules to maintain trust and that internal discipline does not reflect broader safety failures. Middle-ground voices are less certain about the cause of the firings. Tim Duffy suggested the dismissals more likely stemmed from leaking details about an internal model, identified as Astra, to The Information. Garrison Lovely noted that one of OpenAI's own safety leaders strongly agreed with the letter's recommendations while still maintaining the firings were unrelated to raising safety concerns.

The clearest near-term signal to watch is whether OpenAI's board responds to the open letter, and specifically whether the company makes any public commitment on chain-of-thought monitoring or independent audits. Any further departures from the safety team would also shift the weight of the competing accounts considerably.