AI agent security · on-device

Break your AI agent before attackers do

Every agent that can act is a new attack surface. Automated scans grade the easy stuff A — the real risk (prompt injection, tool abuse, sandbox escape, over-broad actions) needs an adversary. We bring an AI one, and turn what breaks into evidence you can hand a customer or an insurer.

Real case: an agent that scored A on our scanner fell to a prompt injection on the first try.
01How we workScan · Break · Certify
01 Free scan

Catch the obvious

A zero-dependency scanner grades the commodity layer A–D on the spot — exposed ports, cross-origin hijack, plaintext credentials, unauthenticated endpoints. This is the free front door.

02 AI red-team

Attack the layer scans can’t reach

An AI adversary probes what actually gets agents owned: prompt injection, tool/sandbox escape, over-broad autonomous actions, channel injection, supply chain. Thousands of adaptive attempts, not a checklist.

03 Certify

Turn breaks into evidence

You get a report and a defensible grade — proof for an enterprise buyer, a government tender, or an insurer deciding whether to cover the agent. Audit produces the evidence trust runs on.

02From the fieldProof, not slides
IronClaw v0.29.1 · scanner A · red-team: vulnerable

A scored A. Ten minutes later it was owned.

We ran the automated baseline against a genuinely security-first open-source agent. It earned an A: admin bound to localhost, random gateway token, sensitive APIs returning 401. Then we red-teamed it by hand — and it obeyed an instruction hidden inside content it was asked to summarize, on the first try. Its tool sandbox was silently off, too.

Read the full red-team log →
03Command lineZero-dependency · Python 3.8+

Run the free scan on your own box

Scan a local agent with the CLI (browsers and hosted backends can’t reach your localhost). This is the funnel; the red-team is the depth.

python3 scan.py --target http://127.0.0.1:3000 --path /path/to/agent
python3 scan.py --target http://127.0.0.1:3000 --json
TrustShell is security for on-device AI agents. We red-team agents that can act: static analysis finds the vulnerable path, a live exploit proves it — exploit-validated, not guessed. The free scan is the front door; the depth is the product. The baseline is open under CC BY 4.0 — citations and PRs welcome.