AI Security Labs
Sharing resources for securing AI systems: practical guides, example configuration to adapt and test before you use it, and writing on how AI tools fail and how to keep them contained.
Upcoming books
Fundamentals of AI SecurityUpcoming
AI Security and Governance: Strategy and LeadershipUpcoming
Microsoft Copilot Studio Governance and Security for PractitionersUpcoming
Syllabus and details for every book
Latest from the blog
- Security for Cursor Development
Cursor's agent runs commands with your permissions. This post covers the risks, the incidents that turned configuration into code execution, their impact, and the settings, checks and habits that control them.
- Security for Gemini Development
Gemini CLI and Code Assist agent mode run commands with your permissions. This post covers the risks, the incidents, their impact, and the settings, checks and habits that control them.
- Security for Copilot Development
GitHub Copilot's agents run commands with your permissions. This post covers the risks, the incidents that turned injected text into commands, their impact, and the settings, checks and habits that control them.
- Security for Codex Development
OpenAI Codex runs commands on your machine with your permissions. This post covers the risks, the CVEs that began with opening a repository, their impact, and the configuration, rules, checks and habits that control them.
- Security for Claude Code Development
Claude Code runs in your terminal with your permissions. This post covers the risks, the 2026 incidents that began when a developer opened a repository, their impact, and the settings, hooks, checks and habits that control them.