Monday AI Radar #13
New models from Anthropic and OpenAI show dangerous capabilities are outpacing safety evaluations.
Deep Dive
The newsletter 'Against Moloch' analyzes rapid AI progress. Anthropic released Claude Opus 4.6, a major upgrade just two months after Opus 4.5. OpenAI launched GPT-5.3-Codex, with a 'Spark' variant processing over 1,000 tokens/second. Experts warn these frontier models are reaching plausibly dangerous capability levels, but safety evaluations are failing to keep pace, leaving the industry 'flying blind' according to Anthropic's own system cards.
Why It Matters
Professionals across law, finance, and tech face imminent, massive disruption as AI capabilities surge ahead of our ability to control them.