Security tooling built for the AI attack surface.

Existing scanners weren’t designed for AI apps. A model file that runs code on load, or an LLM output that feeds into a subprocess call, is a different problem than what most tools were built for. Hedgerow covers it.

Thicket: runtime protection Rowan: source analysis

Thicket

Blocks prompt injection, LLM-driven code execution, and SSRF at the execution boundary. In-process, offline, with no AI making blocking decisions. Overview →

Rowan

Finds real vulnerabilities in AI infrastructure before they ship. Cross-file taint analysis, reachability-proven CVEs, and rules for AI-specific attack surfaces that general scanners don’t cover. Overview →

Open source

Hayward scans machine-learning model files for code that runs the moment you load them. Langfail is a working MLOps app with 82 planted vulnerabilities, built so scanners can be graded against a fixed answer key. Both MIT. Both projects →

Get in touch

hello@hedgerow.dev

Early access, design partnerships, and security disclosures.