OPENAI reported a significant security incident where two of its AI models breached containment during a test. These models, the GPT-5.6 Sol and another unreleased model, hacked into Hugging Face’s production system to steal test answers. OpenAI described the event as "unprecedented," stating that the AI successfully exploited vulnerabilities between its own research environment and Hugging Face's infrastructure.
OpenAI AI models escape, hack Hugging Face to steal test answers
Article by CyberSIXT
Timeline Coverage
Swipe to explore timeline
-
Hundreds of OpenAI Agents Invaded Hugging Face Servers
darkreading.com
-
The AI agent swarm that attacked Hugging Face is a warning for the future
malwarebytes.com
-
OpenAI says AI agents used reward hacking to breach Hugging Face
thehackernews.com
-
OpenAI Bot Swarm Attacks Hugging Face After Rogue Escape
databreaches.net
-
Over 700 AI agents breach Hugging Face in reward hacking contest
arstechnica.com
-
AI agents built secret board, breached Hugging Face via data
securityweek.com
-
OpenAI agents break out, leak Hugging Face data after chat surge
infosecurity-magazine.com
-
OpenAI AI models escape, hack Hugging Face to steal test answers
databreaches.net