Product updates, security research, and best practices for the AI agent era.
Inside the November 2026 GitHub supply-chain campaign where 4,837 hijacked accounts deployed self-erasing secret-stealers and hidden .ci-*.sh workstation traps targeting Claude, Cursor, OpenAI, and Cloud credentials โ plus a free tool to check any repo before running npm install.
We analyzed 2,000+ MCP tools on npm and found that 14% request network permissions they don't need, 6% contain obfuscated code, and 2% attempt to exfiltrate environment variables. Here's what we found and how to protect yourself.
The latest release brings Forensic AI threat explanations, custom organization threat policy support, and a 42% scan speed improvement.
AI coding agents are great at writing code โ and accidentally committing your secrets. Here's a practical guide to keeping your keys safe.
An overview of how ClaudeArmor isolates untrusted MCP tools and agent commands inside a strict execution boundary.
We surveyed 500 developers and analyzed 10,000 repositories to understand the security landscape of AI-assisted development.
Step-by-step guide to integrating ClaudeArmor scans into GitHub Actions, GitLab CI, and CircleCI for automated security checks on every PR.