Mythos Preview is the first AI model purpose-built for offensive security research — discovering novel vulnerabilities, building exploit chains, and reversing closed-source binaries at machine scale. Register your device to unlock your API key and the Enterprise Agent.
Claude Code. Live exploitation. End-to-end autonomous pipelines. One command from binary to working exploit.
Design end-to-end security research pipelines — zero-day hunts, exploit chains, reverse engineering runs, and IR triage. Mythos handles the full workflow.
Mythos Preview was built under Project Glasswing — Anthropic's initiative to find and patch vulnerabilities in the world's most critical software before they can be weaponized. Every offensive capability exists in service of defense.
proof-of-concept.js (shell exploit #181)
security@mozilla.org
Autonomously identifies novel vulnerabilities in production software — operating systems, web browsers, cryptographic libraries, and kernel code. Found vulnerabilities in every major OS and browser tested.
Constructs multi-stage exploit chains end-to-end — JIT heap sprays, CFG bypasses, sandbox escapes, kernel LPEs. Produced 181 JavaScript shell exploits in a single Firefox session vs. 2 for the prior model.
Decompiles binaries and reconstructs source logic to find vulnerabilities in proprietary software — EDR drivers, firmware, DRM systems, and undocumented protocols — without source access.
Converts publicly disclosed but unpatched vulnerabilities into working, reliable exploits — including a 27-year-old OpenBSD SACK vulnerability requiring signed integer overflow chaining across multiple packets.
Expert-level severity assessment — 89% alignment with human security validators on CVSS scoring, risk classification, and exploitability analysis across thousands of high- and critical-severity findings.
Maps attacker TTPs to MITRE ATT&CK, correlates EDR telemetry with memory forensics and network captures, extracts IOCs, and generates full incident reports in minutes, not days.
Mythos Preview API keys are issued only after hardware device verification. Each key is cryptographically bound to a single verified Mac — non-transferable, revocable, and fully audited.
Device registration also activates the Anthropic Enterprise Agent for macOS — a native AI assistant that integrates with your security tooling, ITSM, and dev stack for day-to-day research workflows.
Every Mythos capability is developed under Project Glasswing's defensive mandate — to secure critical infrastructure before attackers can exploit it. Offensive research tools for defensive outcomes.