Anthropic has published an initial report for Project Glasswing, the cybersecurity initiative it launched in April that aims to prevent AI cyberattacks with, well, AI. The initiative is powered by Claude Mythos Preview, the company’s unreleased model, which Anthropic says has already helped its partners find more than ten thousand vulnerabilities overall just a month after Glasswing’s launch. In addition, it says most of its partners have “each found hundreds of critical- or high-severity vulnerabilities in their software” using the model.
The company said that its partners’ rate of bug-finding has increased by more than a factor of ten. Cloudflare found 2,000 bugs, 400 of which are high or critical in severity. Mozilla previously reported that it found and fixed 271 vulnerabilities in Firefox, 10 times more what it found in an older version of the browser using another Claude model.
Microsoft’s recent announcement that its patch releases will “continue trending larger for some time” is apparently because of the bugs it found through Mythos Preview. Anthropic also used Mythos Preview to scan 1,000 open-source projects over the past few months and found 6,202 high- and critical-severity vulnerabilities out of 23,019. While the company didn’t include it in the report, a security research firm recently claimed that it found a way to breach macOS, an operating system known for having tight security, with help from Mythos’ bug-finding capabilities.
The company explained in its report that it hasn’t released Mythos Preview to the public yet, because no company (including itself) has developed safeguards strong enough to prevent models like it from being misused. It intends to release “Mythos-class models” in the future, though, when those safeguards become available. For now, it’s planning to work with partners like the US and other governments to expand the availability of Project Glasswing. That indicates that the company may be on its way to repairing its relationship with the US government. The company is already working with several partners at the moment, including Amazon Web Services, Apple, CrowdStrike, Google, JPMorganChase, NVIDIA and Palo Alto Networks, in addition to the others we’ve already mentioned.
Anthropic is reportedly about to be profitable for the first time since it was founded in 2021. According to a recent report by the The Wall Street Journal, it’s on track to post a revenue of $10.9 billion with an operating profit of $559 million for the quarter ending in June. The company doesn’t expect to remain profitable in the quarters that will follow, however, as it intends to invest more money into computing resources and other expenses.
