An off switch for dual-use knowledge in AI models

Anthropic published a research article discussing an “off switch” for dual-use knowledge in AI models, apparently aimed at limiting the generation or use of potentially dangerous information. The available record is only a Hacker News listing and short summary, so the mechanism, evaluation methodology, quantitative results, and deployment boundaries cannot yet be established. The article may be important for model safety and capability governance, but its technical claims require verification against the original source.