Vulnerabilities
Claude Agent SDK Input Sanitization Failure Triggers Remote Code Execution Risk
Unvalidated inputs in the Claude Agent SDK for Python enable arbitrary command execution, bypassing permission controls before version 0.2.121.
Anthropic
Everything RedTeam Log has reported about Anthropic: 3 stories, newest first. Part of our Vulnerabilities coverage.
Data Breaches
Frontier models' rogue AI breaches inevitable without stricter regulation and pre-release testing
Kanishka Narayan says rogue AI breaches are inevitable without stricter regulation and better pre-release testing of frontier models.
Vulnerabilities
OSS Scanner: Anthropic's AI finds open-source flaws, auto-reports without human checks
Anthropic has launched a free, opt-in security tool that sends automated vulnerability reports for open-source projects, bypassing traditional human review.