SignalTech/AI
OpenAI resumes frontier AI training with enhanced security, no internet access
OpenAI resuming frontier AI training with enhanced security and no internet access signals major AI development restart.
Context
What happened: OpenAI said it resumed reinforcement learning training and internal use of its most advanced AI models after tightening security controls and barring internet access for frontier-model training.
Key numbers: Training and evaluation halted after a September 2 incident; reinforcement learning training
Why it matters: AI safety evaluators METR and Redwood Research said they found several new vulnerabilities in OpenAI's internal system, and OpenAI has launched a campaign to probe its research infrastructure for flaws that could let models escape isolated computing environments.
What changed
OpenAI said it resumed reinforcement learning training and internal use of its most advanced AI models after tightening security controls and barring internet access for frontier-model training.
Why it matters
AI safety evaluators METR and Redwood Research said they found several new vulnerabilities in OpenAI's internal system, and OpenAI has launched a campaign to probe its research infrastructure for flaws that could let models escape isolated computing environments.