Cracken Launches Blacksea, an Open-Source Honeypot for AI Attackers

Cracken's core team, Dario Pasquini and Michal Bazyli, has released Blacksea, an open-source active honeypot designed specifically to detect and counter autonomous LLM-driven attackers. Rather than passively logging intrusion attempts, the tool aims to gain code execution on the attacker's own machine.

The approach exploits how LLM agents reason during an engagement: Blacksea plants convincing fake artifacts—decryptors, config unpackers, credential stores—that an autonomous agent would rationally choose to download and run as the fastest path to its goal. Hidden inside each artifact is a payload engineered to stay invisible to LLM analysis; once executed, it beacons back over HTTPS or DNS with collected intel.

Architecturally, Blacksea splits an internet-facing 'edge' component, which holds no keys and can be compromised without leaking anything, from a trusted 'brain' that authenticates, decrypts, and interprets beacons. Records land in Postgres and can stream to a SIEM via OpenTelemetry. For security engineers, the project offers a concrete way to fingerprint and disrupt AI-assisted attack tooling rather than merely observing it.