Hey Product Hunt 👋 Ivan here, founder of Navil. RSAC 2026 last week made agent security a real category overnight — 10+ vendors launched products. That's great for awareness. But here's what bugged me: every single one protects you in isolation. Your scanner, your rules, your threat data. Nobody else benefits from what you learn. That felt wrong. Security threats against AI agents aren't unique to one company. A prompt injection that hits your MCP server will hit mine tomorrow. A tool poisoning pattern discovered in Cursor will show up in Claude Desktop next week. Why should each of us discover the same attack independently? So we built Navil around a different idea: a community threat pool. Every Navil deployment runs 12 anomaly detectors that learn what "normal" looks like for your specific agents. When something abnormal happens — a novel attack pattern, an unusual tool call sequence, a denial-of-wallet attempt — that pattern gets anonymized and shared with the network. Within seconds, every other Navil deployment can detect the same thing. 568 detection patterns so far. Growing autonomously. No single company could build this catalog alone. The network does it together. The thing I'm most proud of is how easy it is to join: pip install navil && navil secure 60 seconds. Discovers your configs, wraps your servers, generates policies from the community's collective intelligence, shows you a before/after score. Most people go from ~23% to ~78% on their first run — because they're getting the benefit of every detection the network has ever made. We also open-sourced the threat catalog separately: 219 attack vectors, 11 classes, OWASP Agentic Top 10, real CVE references. CC BY-SA 4.0. Useful even without Navil — think MITRE ATT&CK for AI agents. Why open source? Because security tooling should be auditable. And because the community threat pool only works if people trust what's running. Apache 2.0, fully local, no cloud dependency. I'd love to hear: what's the scariest thing your AI agents have done that you didn't expect? That's exactly the kind of pattern the threat pool is built to catch. GitHub: https://github.com/navilai/navil Threat Catalog: https://github.com/navilai/navil... Live Threat Radar: https://navil.ai/radar Blog (8 deep-dives): https://navil.ai/blog

Navil.ai

AI agent security that gets smarter with every deployment

AI agent security that gets smarter with every deployment