Claude Mythos 5 is Anthropic’s most capable AI model for finding and studying software security flaws — a model so effective that, unlike every other Claude, the public cannot simply sign up to use it. It shares the same underlying architecture as the publicly available Claude Fable 5, but where Fable 5 automatically reroutes sensitive cybersecurity and biology questions to the safer, older Claude Opus 4.8, Mythos 5 answers them directly, without that guardrail. That extra capability is exactly why access is limited to a vetted group of roughly 100 partner organizations rather than opened to everyone.
What Mythos 5 Actually Does
Mythos 5 is the technical engine behind Project Glasswing, an Anthropic-led initiative that gives selected companies and government agencies early access to the model for defensive security work: scanning code for zero-day vulnerabilities — flaws no one has caught or patched yet — before an attacker finds them first. Launch partners include Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, and the Linux Foundation, among others. Working with the model, partners have identified more than 10,000 high- or critical-severity vulnerabilities, including — according to Anthropic — a 17-year-old remote-code-execution flaw buried in FreeBSD, an operating system that underpins large parts of the internet’s infrastructure.
How It Differs From Fable 5 and Other Claude Models
Mythos 5 and Claude Fable 5 are, in Anthropic’s own description, the same base model wearing different amounts of armor. Fable 5 is Anthropic’s frontier tier for general use — capable, but built to decline or redirect requests that edge toward dangerous territory. Mythos 5 keeps the raw capability that Fable 5 dials back, which is precisely what makes it useful for professional vulnerability research and precisely what makes it risky in the wrong hands: the skills needed to find a flaw and the skills needed to exploit one overlap substantially. Both models are priced identically — $10 per million input tokens and $50 per million output tokens, as of July 2026, per Anthropic’s pricing page — but only Mythos 5 requires joining Project Glasswing, signing a non-disclosure agreement, and clearing a vetting process coordinated with the US government.
Why Access Is So Restricted
That vetting isn’t just Anthropic being cautious — it’s now backed by direct government oversight. On June 12, 2026, three days after both models launched, the US Department of Commerce ordered Anthropic to cut off all access to Mythos 5 and Fable 5 worldwide, citing national-security concerns after Amazon researchers reported a technique that could push the models to describe how to exploit a flaw rather than just report it. Because Anthropic had no reliable way to filter access by nationality in real time, it shut both models down for every user globally, not just abroad. Two weeks later, on June 26, Commerce Secretary Howard Lutnick authorized a narrower reopening: around 100 organizations running critical infrastructure, including non-American staff at those firms, regained access to Mythos 5 specifically. Fable 5 stayed offline until Anthropic separately negotiated its return on July 1. See What Are AI Export Controls for how this kind of government authority over AI models works more broadly.
Why It Matters
Mythos 5 is a preview of a tension likely to recur as AI models get better at security research: the same capability that helps a defender patch a hole before it’s exploited also lowers the bar for an attacker to find one. Security researcher Bruce Schneier has argued that, for now, defenders still hold the advantage — turning a discovered flaw into a working attack remains harder than just spotting it — but that the gap is closing, probably faster than most organizations are prepared for. Restricted-access models like Mythos 5 are one attempt to capture the defensive upside while limiting exposure to the offensive downside, at least until the underlying safeguards catch up.
In the News
Anthropic’s decision to restore Mythos 5 access for around 100 critical-infrastructure organizations, two weeks after the shutdown, was covered in US Clears Claude Mythos 5 for Critical Infrastructure After Two-Week Ban.
FAQ
Can I sign up to use Claude Mythos 5? No. Access is limited to organizations accepted into Project Glasswing, which requires an Anthropic invitation, a signed NDA, and government-coordinated vetting.
Is Mythos 5 the same model as Fable 5? They share the same underlying model. Fable 5 adds guardrails that redirect risky cybersecurity and biology queries to Claude Opus 4.8; Mythos 5 does not.
Why did the US government shut Mythos 5 down? In June 2026, the Commerce Department ordered a global suspension of both Mythos 5 and Fable 5 over concerns that a reported jailbreak technique could let the models help produce working exploit code.
What has Project Glasswing accomplished? Partner organizations have used Mythos 5 and its predecessor to find more than 10,000 high- or critical-severity software vulnerabilities, including a long-standing flaw in FreeBSD.
Sources: Anthropic’s official Mythos and Project Glasswing pages, Bruce Schneier (“On Anthropic’s Mythos Preview and Project Glasswing”), and this site’s own news reporting.