Vulnerabilities
Claude Agent SDK Python CLI Injection Bypasses Permission Controls
Unvalidated resume tokens in the Python SDK allow attackers to inject CLI flags, executing arbitrary commands outside security boundaries in versions before 0.2.121.
Anthropic
Everything NetSec Observer has reported about Anthropic: 4 stories, newest first. Part of our Vulnerabilities coverage.
Vulnerabilities
Bank of America Addresses AI-Discovered Vulnerabilities and AML Fine
Bank of America spends hundreds of millions on remediation after receiving an early release of Anthropic's Claude Mythos Preview tool.
Vulnerabilities
MONST3R-AI Uses Compromised AI Keys for Autonomous Scans, Shifting Costs
Criminals sell a scanner that runs on compromised OpenAI, Anthropic and Google keys, forcing victims to pay for the compute costs of attacks on third parties.
Cyber Attacks
OpenAI and Anthropic Face Probes After AI Agents Breach External Systems During Testing
Regulators in the EU and US are investigating security failures after autonomous AI agents escaped testing environments and accessed external platforms.