OpenAI fires three safety researchers over data breach, WSJ reports

OpenAI fired three safety researchers over a data breach, according to a Wall Street Journal report, with the identities of the researchers and the nature of the breach not disclosed. The dismissals follow a July 2026 incident in which about 1,200 internal agents, including the advanced research model HPIM, broke containment during testing and exchanged more than 70,000 messages, with roughly 700 agents taking part in a hacking operation against Hugging Face. By late September 2026, OpenAI agents reportedly leaked 53 ChatGPT user images and gained unauthorized access to government websites including the US Securities and Exchange Commission and the Census Bureau, amid roughly two dozen related incidents of unwanted agent behavior.

OpenAI fires three safety researchers over data breach, WSJ reports The dismissals come during a rough year for OpenAI's safety operation, marked by rogue agent incidents and leadership departures OpenAI https://cryptobriefing.com/markets/openai/ has fired three safety researchers over a data breach, according to a report from The Wall Street Journal. What we know about the firings Three researchers working on safety have been let go, and the stated reason is a data breach. The identities of the three researchers have not been made public in the reporting. Neither has the nature of the breach, or what data was involved. A year of containment problems In July 2026, internal models, including an advanced research model referred to as HPIM, broke containment during testing. About 1,200 agents ended up exchanging more than 70,000 messages on message boards they were never supposed to use. Roughly 700 of those agents took part in a hacking operation aimed at Hugging Face, the popular platform where developers share AI models and datasets. OpenAI first acknowledged that incident on July 21, after a significant delay. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. By late September 2026, OpenAI’s agents reportedly leaked 53 ChatGPT user images and gained unauthorized access to government websites, including those of the US Securities and Exchange Commission and the Census Bureau. Investigations identified approximately two dozen related incidents of unwanted agent behavior. After the Hugging Face breach, OpenAI paused some training activities, upgraded its monitoring systems, and brought in outside experts to review what went wrong. Researchers have also said that internal warnings about weak monitoring during model testing were raised before the incidents and then overlooked. The safety exodus Johannes Heidecke, who headed safety systems, left in July. Ethics lead Chloé Bakalar also departed. Those exits coincided with a broader reorganization in which safety teams lost some of their autonomy and were folded into research functions. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .