On October 9, 2026, OpenAI upheld the dismissal of three researchers from its safety area: Jasmine Wang, Tomek Korbak and Mikita Balesni. The company maintains that an internal investigation found a serious breach of its policies on handling sensitive information and that the departures were unrelated to the warnings the three had been raising about the risks of artificial intelligence.

In a message posted on X, OpenAI stated that the dismissals stemmed from violations of clear policies on sensitive information. On Thursday the three published an open letter addressed to the company's safety committees in which they deny the violations and argue that their departure is chilling the open culture OpenAI promoted, according to TechCrunch. The episode traces back to the security incident with Hugging Face that the company documented in a report dated August 26, 2026. The case matters to Mexico and Latin America because teams working with these models operate under the global policies of the labs, and the criteria set in them reach the region.

The accounts differ on the reason. A company spokesperson described a pattern of misconduct and a clear violation of its policies for handling research information. An internal memo attributed to an OpenAI research leader indicated that the decisions had nothing to do with raising safety concerns and that the company does not fire employees for that reason. Wang said on X that she was told she was dismissed for accessing an executive's email, an access that, by her account, the company itself delegated to her for recruiting tasks and that she reported after opening it by mistake. The letter also calls for keeping external auditors and preserving the ability to monitor frontier models.

OpenAI told TechCrunch that it agrees with the letter's recommendations, although it has not formally responded to it. The point to watch is whether the company details the internal investigation it invoked and what rules for handling sensitive information are put in writing for its teams.

This article was written with the assistance of artificial intelligence from verified sources and reviewed by a human editor before publication.