Anthropic's Mythos AI autonomously discovers zero-day vulnerabilities
Anthropic announced Claude Mythos, an AI model capable of autonomously discovering and exploiting zero-day software vulnerabilities faster than human experts. The company limited release to 50 industry partners through Project Glasswing rather than make it publicly available, citing security risks from its advanced hacking capabilities.
What happened
Anthropic announced Claude Mythos Preview on April 7, 2026, an AI model with autonomous hacking capabilities. The company deemed the model too dangerous for public release and rolled out Project Glasswing, a limited release to approximately 50 industry partners.
Mythos is reportedly able to identify and exploit previously undiscovered software vulnerabilities (zero-day vulnerabilities) at speed and scale that surpasses all but the most skilled humans. Anthropic states that Mythos has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.
Reported impact
The UK's AI Safety Institute independently assessed that Mythos's cyber capabilities represent a significant step up in AI-enabled hacking. Experts have warned about the potential for agentic AI systems to shift the balance between offense and defense in cyberspace.
Index analysis
Mythos signals the speed at which new and potentially dangerous AI capabilities are emerging and the challenges of mitigating their risks before they widely proliferate. The model is forcing a reckoning with long unaddressed cybersecurity problems.
The limited release to select partners highlights governance dilemmas around enabling safe adoption of progressively more powerful models, including geopolitical tensions that arise when vastly superior models are made available to only a select few.
Governance controls
Anthropic withheld public release of Mythos and instead deployed it through Project Glasswing, a controlled release to approximately 50 industry partners intended to harden cyber defenses against future AI-enabled hacking.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.