Anthropic, the $380 billion AI big liable for the Claude device, has a brand new AI mannequin referred to as Mythos that might turn into a crypto hacking nightmare.
Involved about world panic if it have been to launch its frontier mannequin too quickly, Anthropic handed early “Mythos” entry to JPMorgan Chase, Apple, Microsoft, and some dozen different blue chip tech corporations.
Sadly, Anthropic didn’t grant visitor checklist entry to any crypto firm for its paternalistic Undertaking Glasswing.
One Bitcoin developer requested Anthropic straight, “Why not cooperation with bitcoin/crypto projects?” Anthropic declined to answer.
JPMorgan Chase acquired a head begin. Crypto didn’t.
Introducing Undertaking Glasswing: an pressing initiative to assist safe the world’s most crucial software program.
It’s powered by our latest frontier mannequin, Claude Mythos Preview, which may discover software program vulnerabilities higher than all however probably the most expert people.https://t.co/NQ7IfEtYk7
— Anthropic (@AnthropicAI) April 7, 2026
A wing of glass stopping Mythos from hacking crypto
Anthropic’s Glasswing cybersecurity dash is allowing 50-60 corporations early entry to its unreleased mannequin that “can find software vulnerabilities better than all but the most skilled humans,” in accordance with the corporate.
It is usually donating $4 million value of AI credit and is “committing” as much as $100 million in AI credit for Glasswing.
In response to Anthropic, which clearly has an incentive to reward the powers of its unreleased mannequin for media and fundraising functions, Mythos has sturdy reasoning and coding expertise and is significantly extra harmful as a software program hacking device than most human builders.
Anthropic claims Mythos is “very autonomous” and has already discovered “thousands of high-severity vulnerabilities,” together with bugs in “every major operating system and web browser.”
It withheld particulars about most of these bugs, besides a 27-year-old bug in OpenBSD Unix software program and a 16-year-old flaw in FFmpeg video software program.
Curiously, Anthropic has printed professionally staged and videotaped promotional supplies through which Anthropic stakeholders sound alarms about Mythos’ capabilities.
For media functions, it fastidiously chosen the identify “Glasswing,” which refers back to the Greta oto butterfly whose clear wings resemble glass.
Crypto, excluded from Glasswing, is especially weak
If Mythos’ risk is actual, crypto software program is especially weak to hackers with entry to it.
Many implementations of crypto software program are largely or totally immutable, comprise super monetary worth, and have globally distributed deployment and improve cycles that forestall a fast protection.
Protos reported in December 2025 that even earlier than Mythos, Anthropic had pitted its AI brokers in opposition to 405 sensible contracts.
Even with backdated data and no web entry, its brokers accurately predicted tens of millions of {dollars} value of accessible exploits on sensible contracts which had gone dwell after researchers reduce off the AI web and data entry.
Anthropic’s AI brokers additionally uncovered novel zero-day vulnerabilities in hundreds of contemporary contracts with no beforehand recognized flaws.
These discoveries have been earlier than Mythos. In response to Anthropic’s self-aggrandizing claims yesterday, Mythos can dramatically out-codes all the pieces Claude has constructed beforehand.
Stifel analyst Adam Borg is satisfied. “We read this as having the potential to become the ultimate hacking tool, and one that can elevate any ordinary hacker into a nation-state adversary,” Borg wrote about Mythos.
Anthropic says Glasswing companions will share their findings with the broader business and patch main bugs previous to the general public launch of Mythos.
