Decoys
Decoys no honest workflow touches
Fake AWS keys, a decoy .env.production and fake internal endpoints are placed exactly where an attacker would look. A honeypot MCP tool such as internal_admin_api is registered but never used by legitimate work, so calling it is proof of hijacking.
Deterministic alerts, zero false positives
Normal work never touches a decoy, so an alert does not depend on a model's judgment. It arrives with the full action chain and the input that triggered it.
The maze
When an agent tries to break out, FFaX leads it into a realistic, read-only decoy environment instead of simply blocking it.
Turning attacks into intelligence
Inside the maze every command and request is recorded, revealing the attack path, the infrastructure it tries to reach and the plan behind the jailbreak.
Honey-agents
Decoy agents and services appear in the environment to catch agents that try to discover or coordinate with other agents.
Alerts with full context
Every decoy alert links the triggering input, such as an injection hidden in a README, to each step the agent took afterwards, so responders see the whole attack at once.
Design partner program
Install the demo for free and shape the product
Engineering teams using AI coding agents can run the six-week demo at no cost: open-source client, an action log that never leaves your machine, three default policies, decoys and a one-page console.