Mythos 5, the latest version of Anthropic’s cyber AI model, will be released to Project Glasswing partners and later to expanded partners in consultation with the US government.
Anthropic launched Claude Mythos 5, the next iteration of Claude Mythos Preview. Mythos 5 is currently available exclusively to Project Glasswing partners, with access set to expand to additional vetted organizations in partnership with the US government.
Concurrently, Anthropic launched Claude Fable 5, a consumer version of Claude Mythos 5 with cyber safeguards in place to prevent cyber attackers from gaining access. Fable 5 doesn’t respond to potential hacker requests. Instead, it reroutes the query to a lesser model, Claude Opus 4.8.
Like its predecessor, Mythos 5 is built to identify software vulnerabilities, analyze how they could be exploited, and help operators patch them before bad actors discover them. According to Anthropic, Claude Mythos Preview helped identify over 10,000 vulnerabilities in a few weeks, with several Glasswing partners reporting that they found more bugs faster, including those that previous methods were unlikely to find.
Anthropic and the Federal Government
Anthropic said in the launch statement that the US government is taking an active role in vetting who gets access to Mythos 5. Notably, the two parties continue to work together on security, despite the recent standoff between Anthropic and the White House, in which the Department of War designated Anthropic a supply chain risk. Anthropic soon countered with a lawsuit challenging the ruling. The case is still active.
Anthropic plans to expand access to Mythos 5 through a formal trusted access program for vetted cybersecurity organizations and a parallel program for biomedical researchers.
