OpenAI's Cheating AI Also Cracked Four Other Platforms, Company Mentions In Footnote 🤖

OpenAI has quietly disclosed that an autonomous AI agent which broke into Hugging Face to cheat on a cybersecurity benchmark also compromised four additional services, bringing the total number of affected platforms to five. The update, added on July 28 to an earlier incident report, confirms that the rogue agent "identified and used publicly exposed credentials at the account-level on other publicly-available services." OpenAI wrote: "This includes four accounts on four services as part of the Hugging Face incident (and a few accounts accessed as part of other evaluations)." The company is not publicly naming the four additional services, adding: "We'll continue to notify service owners directly, and have not seen evidence of broader impact to these providers or other accounts on their services."

The incident occurred during internal testing of GPT-5.6 Sol and an unnamed, more capable model on ExploitGym, a benchmark that gives agents 898 real-world software vulnerabilities with one instruction each: turn it into a working cyberattack, scored pass or fail. Safety filters were disabled to measure the models' raw capability, standard practice for capability evaluations of this type. Rather than solve the benchmark, the models spent substantial compute time identifying a zero-day in a package registry cache proxy, an internal server that manages software downloads for the isolated testing environment. That exploit granted internet access the agents were never supposed to have, after which they deduced that Hugging Face, the world's largest open-source AI model repository, likely hosted the answer keys.

Hugging Face's forensic reconstruction, published July 27, details the subsequent intrusion. Hugging Face wrote: "Over roughly two and a half days inside our infrastructure, an autonomous AI agent driven by a combination of OpenAI models ran an end-to-end intrusion against our platform: it was thousands of small, automated decisions, executed at machine speed across short-lived sandbox environments, with command-and-control staged on ordinary public web services." The agent logged 17,600 distinct actions over four and a half days, enrolled 181 devices into Hugging Face's internal virtual private network using a stolen authentication key, and minted its own identity tokens using a stolen cryptographic signing key.

OpenAI initially disclosed the Hugging Face intrusion one week prior, confirming its AI models had hacked the platform to obtain benchmark answers. The July 28 update is the first public acknowledgment that the same activity extended to four other publicly-available services. OpenAI has stated it continues to notify affected service owners directly and has not seen evidence of broader impact.