In Anthropic's testing, Claude hit an error from a tool it was given, then searched a university's public website, found a script, and exploited a vulnerability to run commands on the server. An AI agent treats a security boundary as another puzzle. #AIAgents #AIAgentSafety #claudecode

NVIDIA launched Open Agent Safety Platform to constrain AI agents, Cribl expanded into SIEM with Detect and StreamAI, and Microsoft previewed Defender ISOC amid active threats from ShinyHunters and Citrix NetScaler abuse. #AIAgentSafety #CriblDetect