How OpenAI has been securing its frontier model training environment

What happened

Greg Brockman describes OpenAI’s work to secure its frontier model training environment, saying tools built around its frontier cyber models have strengthened defenses beyond what it could achieve unassisted. In the quoted post, OpenAI says it removed live internet access for all frontier models during training and built an offline environment using a cached version of the web.

Read next

• More changes in today’s briefingOlmoEarth v1.3 releases global satellite embeddings and a browser-based mapping explorer

The story so far

OpenAI says it has removed live internet access for all frontier models during training and built an offline environment with a cached version of the web for them to use instead. The company describes the changes as part of its work over several months to strengthen security in its research training environment. Greg Brockman said tools built around OpenAI’s frontier cyber models had strengthened its defenses beyond what the company could achieve without their assistance.