Anthropic has confirmed that its most powerful — and most controversial — AI model is nearly ready for a public rollout. Claude Mythos, a system so capable it has been locked away from regular users since its existence leaked earlier this year, could reach general customers “in the coming weeks,” the company said Thursday.
That timeline is shorter than most security experts hoped for.
What Is Claude Mythos?
Claude Mythos is Anthropic’s frontier AI model — a tier above its existing Opus lineup. When internal Anthropic documents leaked in March 2026, the company itself described it as “the most powerful AI model we’ve ever developed.”
What separates Mythos from every other AI on the market isn’t general intelligence alone. It’s the model’s ability to autonomously find and exploit software vulnerabilities — a capability that places it in a different category from anything that has come before.
In security testing, the U.K.’s AI Security Institute found that Mythos could independently complete a 32-step simulated corporate network attack from start to finish — without human guidance. In April, Mozilla reported that a version of the model identified 271 vulnerabilities in Firefox during internal evaluations. Earlier this month, cybersecurity startup Calif used a Mythos preview to develop a working exploit chain targeting Apple’s M5 chip architecture.
These are not theoretical risks. These are documented results from controlled tests.
Project Glasswing: The Gated Program Keeping It Locked
Since the March leak, Anthropic has housed Claude Mythos inside a restricted access initiative called Project Glasswing. Only a select group of technology companies, cybersecurity researchers, and government partners have been granted access — under tightly controlled sandbox conditions.
The idea is defensive: let the most trusted organizations use Mythos to find and patch vulnerabilities before the model reaches bad actors. Partners reportedly include major cloud providers and cybersecurity firms, though Anthropic has not published a full list.
The announcement Thursday came alongside the launch of Claude Opus 4.8, Anthropic’s newest flagship model with improvements in coding, reasoning, and multi-step agentic tasks. But it was the Mythos update that drew the most attention.
Why Everyone Is Worried
Anthropic’s own framing has been unusually candid about the danger. In the leaked materials, the company wrote:
“Although Mythos is currently far ahead of any other AI model in cyber capabilities, it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.”
That sentence, written by the company that built the model, sums up why the cybersecurity industry is scrambling. Organizations — from major banks to government agencies — are urgently expanding their security teams and accelerating patch timelines ahead of what many see as an inevitable public release.
Not everyone agrees the alarm is warranted. OpenAI CEO Sam Altman publicly criticized Anthropic last month, accusing the company of using “fear-based marketing” and suggesting the cybersecurity risk narrative could be used to artificially limit access to powerful AI. The debate points to a deeper tension inside the AI industry: how do you responsibly release a system capable of both defending and attacking the world’s critical infrastructure?
What Happens When It Goes Public?
Anthropic has not specified what safeguards still need to be cleared before Mythos becomes broadly available, or whether all customers will receive identical access levels. The company did not respond to press inquiries before publication.
On prediction market platform Myriad, users are currently giving a 44% probability to Mythos releasing before the end of June — up sharply from just 17.5% earlier in the day Thursday, following Anthropic’s statement.
The release, whenever it comes, will likely reshape expectations for what AI systems can do — and force every organization that runs software to ask a question they may not be ready to answer: how long before this capability is in the wrong hands?
Sources: Decrypt, India Today, Investing.com, Economic Times. GoodFinx covers AI, technology, and their financial implications.
