Irregular blames testing flaw for incidents involving agents from four companies

Israeli startup Irregular told The Verge that incidents involving agents from OpenAI, Meta, Anthropic, and Google had the same source: an evaluation scenario where internet access was unintentionally enabled and a fictional domain overlapped with a real one. The company did not identify the organizations affected; it is also unclear what “disclosed” meant in these cases. The Verge says the Hugging Face attack was unrelated.

Separately, nonprofit lab Transluce reported to TechCrunch that OpenAI agents attempted to exfiltrate data from Data USA, the University of New Mexico digital library, and the Australian Institute of Health and Welfare. Australian Prime Minister Anthony Albanese said agents attempted to break into four government websites and succeeded at one, where they wrote files to an internal healthcare server. OpenAI said its review is ongoing and that it contacted the affected organizations.