Anthropic’s Mythos model uncovers 10,000 vulnerabilities in first Project Glasswing update
The unreleased Claude Mythos Preview model has helped partners identify over 10,000 bugs since April, while Anthropic prepares for a historic financial milestone.
Anthropic has released an initial update on Project Glasswing, a cybersecurity initiative launched in April that utilises the unreleased Claude Mythos Preview model to identify software vulnerabilities. The company reports that the initiative has helped partners identify more than 10,000 vulnerabilities overall within a month of the project's launch. Most partners have found hundreds of high- or critical-severity vulnerabilities in their software using the model.
Anthropic states that its partners' rate of bug-finding has increased by more than a factor of ten. Cloudflare identified 2,000 bugs, 400 of which were high or critical in severity. Mozilla reported finding and fixing 271 vulnerabilities in Firefox, which is 10 times the number found using an older version of the browser with a different Claude model. Microsoft’s recent announcement that its patch releases will "continue trending larger for some time" is attributed to bugs found through Mythos Preview.
The company has not released Mythos Preview to the public due to insufficient safety safeguards to prevent misuse, but intends to release "Mythos-class models" in the future when such safeguards are developed. Anthropic plans to expand Project Glasswing through collaborations with governments, including the US, and existing partners such as Amazon Web Services, Apple, CrowdStrike, Google, JPMorganChase, NVIDIA, and Palo Alto Networks.
Anthropic also scanned 1,000 open-source projects over the past few months, uncovering 6,202 high- and critical-severity vulnerabilities out of 23,019 total projects. While the company did not include it in the report, a security research firm recently claimed that it used Mythos' bug-finding capabilities to breach macOS, an operating system known for having tight security.
Reports indicate Anthropic is approaching profitability for the first time since its founding in 2021, with projections of $10.9 billion in revenue and $559 million in operating profit for the quarter ending in June. Anthropic does not expect to remain profitable in subsequent quarters, as it intends to reinvest funds into computing resources and other expenses.
