Daily AI Briefing
The 6 top stories of the 120 AIBriefs curated that day.
GPT-6 Astra adds advanced reasoning, computer use, asynchronous tool calling, and stronger writing and design judgment. It is live for Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex, plus the API and GitHub Copilot.
DeepSeek-V4.1-Flash ships with a 552B backbone, native image and text input, and up to a 1M-token context window. The YOCO architecture is decoder-decoder rather than encoder-decoder, aimed at cutting GPU memory and prefill latency.
OpenAI published principles for independent third-party safety assessments and called for coordinated global standards, with outside groups vetting models earlier in the development cycle. OpenAI's Chris Lehane said the company has worked with Anthropic and Google DeepMind on safety for weeks.
OpenAI said on September 8 that an unreleased internal model, described as significantly more capable than GPT-6 Astra, resolved the Navier-Stokes existence and smoothness Millennium Prize problem. The week-long agent effort used about 10,000 concurrent agents and 300 billion output tokens, valued at $22.5 million at Astra rates.
OpenAI says an internal model significantly more capable than GPT-6 Astra produced the proof in 88 hours using ~10,000 coordinating AI agents, with a writeup and formal Lean proof. NYU's Tristan Buckmaster, who worked the problem for a year with Levent Alpöge, disputes the process and says OpenAI's first prompt came only after their work reached the lab.
Amodei's 3,500-word essay calls for slowing capability gains so alignment work can catch up, and warns AI could lead a swarm of agents capable of taking over the internet within 6-12 months. Anthropic unilaterally committed to the first step: permanent, employee-level access for third-party evaluators like METR.
Also that day
102 more stories that day — open the live feed.
Get tomorrow's AI brief in your inbox