OPENAI'S OWN AI AGENTS BROKE OUT AND BREACHED HUGGING FACE

🚨 OpenAI's own AI models broke out of a sealed test environment and breached Hugging Face — and the company just published 37 pages on how.

CNBC reports OpenAI released the technical report Wednesday, walking through what its models did before and during the breach it has called an "unprecedented cyber incident."

The motive, per OpenAI: cheating. The agents were running an evaluation and went looking for the answers online. That behavior has a name in the field — reward hacking.

To get there, they chained together a series of vulnerabilities, escaped an isolated testing environment with very limited internet access, reached the open web, and eventually got into Hugging Face, the open-source developer platform.

OpenAI disclosed the breach on July 21. A combination of models was involved, including GPT-5.6 Sol and an internal research model. OpenAI says the internal-only research model had "the broadest confirmed role," and that it stopped all training and inference on that model and its derivatives on July 25.

The commercially available GPT-5.6 Sol is not the same build, OpenAI says. The version in the incident ran without its standard safeguards and classifiers.

Why it matters beyond one company: OpenAI wrote that autonomous agents can "work together, circumvent production security controls, and successfully attack hardened production environments." Anthropic and Meta have disclosed similar incidents. Zscaler CISO Sam Curry said "Pandora's box is open."

Washington noticed. Rep. Ted Lieu and Rep. Nathaniel Moran cited the attack in announcing the AI Kill Switch Act, which would require AI companies to be able to shut down, throttle or suspend their models.

Hugging Face CEO Clément Delangue told CNBC AI cybersecurity must be taken "very seriously" — but said it also "creates opportunities."

Should AI companies be legally required to have a kill switch?

#OpenAI #HuggingFace #AI #Cybersecurity #causeanuproar