AI safety, alignment, red-teaming, jailbreaks, and interpretability.
Mon, July 6Open Source
The GitHub trending project agentskills/agentskills provides specifications and documentation for Agent Skills, gaining 2262 stars in 7 days.
Fri, July 3Research
Anthropic discloses the web security mechanisms and jailbreak framework details of Fable 5.
Thu, July 2Model/API
Anthropic redeployed Claude Fable 5 on July 1 after US export controls were lifted, and added a safety classifier.