OpenAI and Anthropic this week
Astra for Law, misalignment reports, one Claude, Hacktron, standards body (Week 38, 2026) Hacktron hacked OpenAI in late July under OpenAI's bug bounty safe harbor, using Claude Opus 5 to exploit a Discourse bug and grab ChatGPT and Codex sign-in
Development progress of this Othello AI Black is a level 1 AI, white is a level 10 AI The depth of reading is up to 16 moves ahead for both
More than 90% of the development was done by Claude Opus 5.
After using ChatGPT, I learned to touch grass while it reasoned.
I've spent the last 2 hours in stealth building a new type of frontier AI model that I am releasing today: BEV • 20-200x slower than JEV • As cheap as JEV • Matches frontier response times (Opus 5, Sol)
You have reached the end of the archive
All of Claude Opus 5