ai-security
When Anthropic Talks About Its GLM Rival: The Cyber AI Line Has Moved
Anthropic says rival open-weight model GLM-5.3 can build end-to-end exploits and has safeguards that are easy to bypass. What the tests show and what defenders should do.