Greg Brockman @gdb

A post on how we've been securing our frontier model training environment. We've been building tools around our frontier cyber models, which have secured us far beyond what we could have unassisted and should generalize as key to how defenders secure themselves going forward. [引用 @OpenAI]: Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications. On September 20, after a model exploited a DNS misconfiguration…

• What Is AGI? OpenAI's Brockman Says GPT-6 Astra May Already Be There2026-10-28
• Data center backlash becomes a major 2026 election issue2026-10-27
• OpenAI's Brockman Withdraws Second $25M Donation to AI Super PAC2026-09-30
• OpenAI Has Gone Nine Months Without a Communications Chief2026-09-28
• OpenAI's Brockman Optimistic About AI2026-09-17