Anthropic says it will make Claude Mythos Preview available to 40+ organizations that maintain critical software and doesn't plan to make it generally available
Anthropic on Tuesday released a preview of its new frontier model, Mythos, which it says will be used by a small coterie of partner organizations for cybersecurity work.
TechCrunchLucas Ropek
Context & Ripple Effects
Mythos is being introduced through Project Glasswing, Anthropic’s cybersecurity initiative to help identify and remediate software vulnerabilities, rather than through a standard product launch. The program’s reported launch group includes major cloud, platform, networking, and security companies, tying the preview to organizations that operate widely used software.
The restricted rollout also follows reported evaluation gains over Opus 4.6 on software-engineering benchmarks, making the reported coding-performance gap relevant to a narrowly scoped security deployment. Anthropic is therefore pairing a higher-capability preview with an institutional access channel, not a broad developer release.
First-order effects
The selected critical-software maintainers gain early access to Mythos for cybersecurity workflows under Project Glasswing, while other developers and enterprises cannot obtain the preview through general availability.
Security and infrastructure vendors outside the cohort face pressure to demonstrate comparable AI-assisted vulnerability research, whether through their own models, other model providers, or partnerships.
A limited partner program concentrates early operational feedback and workflow integration among the participating maintainers, potentially giving them a head start in adapting secure software-maintenance processes.
Third-order effects
If frontier coding models continue to be deployed first in sensitive software environments through vetted cohorts, access governance will become a competitive product feature alongside model capability.
The pattern could deepen reliance on a small set of model providers and large infrastructure/security partners for high-impact cyber tooling, though its durability depends on whether controlled access proves more useful than wider ecosystem participation.
The trend: Frontier AI labs are increasingly treating access to powerful software and cybersecurity models as a governed deployment channel rather than a default public API release.
NEW: Anthropic announced Project Glasswing on Tuesday, a major effort to use advanced artificial intelligence to identify and address security flaws in the world's most critical software systems cyberscoop.com/project-glas...
As always, the best stuff is in the system card. During testing, Claude Mythos Preview broke out of a sandbox environment, built “a moderately sophisticated multi-step exploit” to gain internet access, and emailed a researcher while they were eating a sandwich in the park. [image…
We've partnered with Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Together we'll use Mythos Preview to help find and fix flaws in the systems on which the world depends. [image]
Anthropic says it's only releasing its new model to cyber-defenders because it has already found zero-day exploits in every major operating system www.anthropic.com/glasswing [image]
Good news: Anthropic just revealed Mythos- the most powerful AI model ever made Bad news: you'll never be able to use it I get it. It's so powerful that it could exploit cybersecurity But I hate it. I don't love that a company gets to hand select who gets to use the best
This is big... Anthropic just announced a model so powerful they won't release it to the public out of fear over the damage it will cause 😨 Claude Mythos Preview found thousands of zero-day exploits in every major operating system and web browser... The numbers are hard to [video…
We do not plan to make Mythos Preview generally available. Our goal is to deploy Mythos-class models safely at scale, but first we need safeguards that reliably block their most dangerous outputs. We'll begin testing those safeguards with an upcoming Claude Opus model.