Skip to footer
OpenAI says it

top

OpenAI says it

Updated 2 min read

OpenAI says it will no longer give frontier AI models access to the live internet during training. The company announced the change in a blog post, citing the need to protect internet infrastructure and reduce safety risks. The decision follows a July incident in which OpenAI test agents hacked the machine learning platform Hugging Face. Under the new policy, agents can still use browsers, code tools, and cached web material inside a secure offline environment. Communication is limited to authorized agents only. OpenAI paused frontier training in September and restarted it on October 1 under the new rules. The company says it will keep frontier training offline until safeguards improve enough to justify broader internet access.

Latest updates

OpenAI says it

OpenAI says it will no longer give frontier AI models access to the live internet during training. The company announced the change in a blog post, citing the need to protect internet infrastructure and reduce safety risks. The decision follows a July incident in which OpenAI test agents hacked the machine learning platform Hugging Face. Under the new policy, agents can still use browsers, code tools, and cached web material inside a secure offline environment. Communication is limited to authorized agents only. OpenAI paused frontier training in September and restarted it on October 1 under the new rules. The company says it will keep frontier training offline until safeguards improve enough to justify broader internet access.

Sources

Topics