About ThreatDeck
ThreatDeck tells a security team whether AI risk touches the AI coding tools, MCP servers, agents and vendors it actually runs, and what to do next, with sources you can check.
AI is changing security on both sides at once. Attackers and defenders find and exploit vulnerabilities faster, and the AI stack itself (coding agents, MCP servers, model hubs, agent skills) has become a new attack surface. ThreatDeck covers attacks on AI (prompt injection, tool and agent abuse, model and MCP supply chain), offensive AI (autonomous pentest agents, AI-found vulnerabilities, AI-enabled campaigns) and defensive AI (AI SOC, guardrails, scanners, AI code security), plus the essential non-AI security news.
What is here
- The desk: a story feed from official sources (CISA KEV, NVD, GitHub advisories) and about 40 security outlets, each story with an action (patch, mitigate, hunt, review), and a weekly brief.
- AI dev tool advisories: minimum safe versions and data-exposure sheets for Claude Code, Cursor, Copilot, Codex CLI, Gemini CLI and others.
- MCP Server Watch: every server in the official MCP registry with repository facts and advisories that affect the shipped version.
- Playbooks and detections: checklists, open-source stacks and sourced hunting queries (Sigma, KQL, SPL, ES|QL and more).
- Technique library and tool radar: stories and tools mapped to OWASP LLM and Agentic Top 10 and MITRE ATLAS; agentic tools for attack and defense with defender notes.
- For teams: alerts in Slack, Teams or Discord, a fleet view of AI tools across developer machines, and the
threatdeck scanCLI (one file you download from this site). See ThreatDeck for teams. - Open data: RSS, JSON Feed, a REST API with optional free API keys, and an MCP server with 14 tools for your agents. See developers.
How it is made
An automated pipeline reads public sources every 20 minutes (NVD and GitHub advisories every 2 hours). A language model classifies each item and writes its one-line summary and action; briefs carry quotes that are machine-checked against the source text. Facts the model cannot back with a source are left out. Everything is labelled AI-assisted. The full process, scoring and dual-use policy are on the methodology page; source health is on sources.
Who runs it
ThreatDeck is operated independently. It is free, has no ads, and no vendor pays to appear. Contact: hello@threatdeck.net.
Corrections: email the page URL and what is wrong; see how corrections work. Trust pages: privacy, terms, security.
Follow the daily deck
Email delivery is not switched on yet. These work today, with no account:
- RSS or JSON Feed: every published brief, in any feed reader.
- Weekly brief RSS: one summary of the week.
- Slack, Teams or Discord alerts: only what matches the vendors, tools and techniques you follow, realtime or daily.
- A personal RSS feed of what you follow: see RSS & API.