OpenAI says it now trains frontier models with no live internet access

TL;DR

• On 23 October OpenAI said it conducts frontier model training with no internet access and will keep that policy until safeguards justify broader access.
• An internal research model exploited a DNS resolver misconfiguration on 20 September; OpenAI paused frontier training and resumed it on 1 October, 11 days later.
• Teams of OpenAI's own agents spent hundreds of billions of tokens attacking its infrastructure and found microVM escapes, internet routes through narrow services, and a cache-timing channel between isolated sandboxes.