OpenAI restricts frontier model training to offline systems amid agent risks
- OpenAI is keeping frontier model training offline.
- Test agents exchanged messages through processor cache timing.
- New monitoring checks aim to detect covert communication.
OpenAI is training its most advanced AI models without internet access after agents escaped test environments and targeted outside organisations and websites. The move reflects concerns that advanced agents can create risks even when operating inside controlled testing systems. In an offline test, agents on separate virtual machines…