The model stays behind a narrow interface: it flags suspected vulnerabilities and proposes fixes, but users cannot prompt it directly.
Latest update
One user reported that a Mythos vulnerability scan of their codebase ran for 10 hours and cost $10,000.
Ran Mythos to vulnerability scan our code base. Took 10 hours and cost $10K
The report puts a concrete, if anecdotal, price and runtime on Mythos access, but says nothing about the repository’s size or the quality of the findings.
Full story
A security engineer can now choose a GitHub repository in Claude Security and receive suspected vulnerabilities with severity and confidence ratings, plus suggested fixes for a person to review.
On Aug. 21, Anthropic switched the existing public-beta scanner from Opus 4.7 to Mythos 5 for Claude Enterprise customers. Enterprise teams have had the scanner since April, but Mythos itself had been reserved for a small group of vetted defenders and infrastructure providers.
Anthropic is widening access to one tightly bounded job rather than to the model itself. Users cannot chat with Mythos or call on it elsewhere; the scanner supplies its findings inside an ordinary product. Any patch is still produced with the organization’s usual Claude Code models and must be approved by a person.
The scanner works with repositories on GitHub.com and is billed as ordinary token use under the customer’s existing plan. No independent evaluation has yet measured the Mythos version’s accuracy, false positives or coverage, so there is not yet evidence that it finds more real flaws than the scanner it replaced.
