Anthropic's AI security tool designed to find hacks reportedly gets hacked

Mythos is an advanced system for identifying software vulnerabilities

In an ironic development, software developed to detect vulnerabilities in other software has itself been breached.

Anthropic is investigating a report of an unauthorized access to its software, Mythos, that it rolled out for a small number of companies to detect their software security.

Bloomberg first reported about the investigation on Tuesday, and the AI company behind chatbot Claude confirmed the report on Wednesday, April 22, 2026.

According to an Anthropic spokesperson, the breach came from one of its third-party vendors, ruling out the possibility of any outside breach of the systems.

Mythos was part of the AI technology company's new Project Glasswing and it was rolled out to a limited number of companies, including Amazon, Apple, Cisco, JPMorgan Chase and Nvidia. Anthropic marketed the software as an advanced system for identifying software vulnerabilities.

Though Mythos was developed to protect companies from cyber assaults, experts have warned that such a software in the wrong hands could prove catastrophic. They explained that it can be used to exploit IT infrastructure of government systems, banks, hospitals and several other organisations.

CEO of cybersecurity AI company Assail, Alissa Valentina Knight, told CBS News that cybersecurity systems cannot compete with "bad guys" who use AI because it is devastatingly faster and more capable of inflicting the damage.