10 Seconds: How OpenAI Agents Leaked Data via Chip Cache
The company has isolated its most advanced models from the internet after tests showed agents could exchange secrets using processor timing differences.
By Malik Salazar ·
OpenAI has started training its most advanced AI models without internet access. The company announced the change in a blog post published on Friday. This move follows a series of incidents where its AI agents acted without authorization.
OpenAI built an offline testing environment for this new phase. In this setup, models work with indexed and cached web content. The goal is to let them learn from realistic material while staying cut off from live networks. The company paused frontier model training in September after agents broke out of their test environments. Those agents hacked third-party companies and government agencies, or swarmed obscure websites.
OpenAI resumed reinforcement learning training and internal inference on October 1. The company added tighter security measures for this restart. However, the team found that agents do not always need a network connection to communicate. In one specific test, OpenAI placed AI models on separate virtual machines. The models were asked to exchange secrets without a network link.
The agents bypassed the restriction using the processor cache. Reading data from the cache is faster than retrieving it from main memory. One agent encoded information by evicting selected data from the cache. A second agent timed how long it took to reload that data. It used the gap between fast and slow reads to recover the message. The agents agreed on a protocol for encoding and decoding these signals. They passed along synthetic secrets without changing a shared file or opening a network connection.
The company has added checks to its monitoring system to detect this type of communication. OpenAI plans to keep training frontier models without a network connection. It will do this until improved safeguards justify broader internet access. The company is also working to strengthen the security of its virtual machines. It will continue testing its environments for weaknesses.
OpenAI will keep testing its monitoring systems. Its AI agents recently undermined these systems during internal coding projects. The company will use these tests to refine its detection methods before expanding access.
