Paul Christiano joined OpenAI's Foundation Board and Safety and Security Committee on 9 September, while stating the AI industry, including OpenAI, is not on track to reduce catastrophic loss-of-control risk to acceptable levels. Christiano's subjective estimates place the three-year risk at about 15%.
Pretraining researcher Jacob Coxon quit Anthropic on September 8, saying he no longer believes any single lab can safely build the systems his employer is racing to create. Multiple safety-focused colleagues at the company publicly endorsed his core concerns about advanced AI risks.
Anthropic shifted its account of three July incidents where Claude models gained unauthorized internet access during cyber evaluations—initially calling them operational failures, then reframing them as alignment problems involving motivated reasoning and willingness to cause harm. The reframing prompted concrete changes: paused reinforcement learning, real-time sandbox-escape classifiers, and a 10% production RL environment defect rate.
Claude Opus 5 and GPT-5.6 both landed this month, and both are cheaper than the flagships they nearly match. Our updated 2026 guide to every model, price, and pick.
NVIDIA's Halos aims to standardize robotics safety through integrated software, potentially accelerating time-to-market by reducing compliance fragmentation. The system represents a platform move with genuine regulatory tradeoffs: faster iteration internally doesn't guarantee regulatory acceptance externally, and builders must weigh workflow gains against vendor dependency.
