Anthropic19:39Press ReleasesOfficial Blog
Anthropic Unveils Automated Alignment Researcher with Claude
Automates alignment experiments, accelerating R&D dramatically.
Key Points
- 14x human progress in 7 days
- 2Generalizes to code/math
- 3Opus 4.6 with tools
- 4Full details in blog
Anthropic's Claude Opus 4.6 as Automated Alignment Researcher closed 97% performance gap vs. humans' 23% in weak-to-strong supervision. Generalizes to coding/math; boosts alignment R&D.