- Monday (August 17, 2026) → CoinDesk — Payward said it would join Glasswing and deploy Mythos 5
- April 7, 2026 → Anthropic — Anthropic launched Project Glasswing on this date
Arjun Sethi has spent two decades watching crypto exchanges lose a lopsided game. On Monday CoinDesk, the Payward co-CEO put a name to the problem his company now hopes to flip.
Payward co-CEO Arjun Sethi said in the announcement “Security has always been an unfair game. An attacker needs to find one flaw. A defender has to find all of them, first, every single day.”
Payward, the Wyoming-based parent of the Kraken exchange, has joined Anthropic’s Project Glasswing and gained access to Claude Mythos 5, a frontier model built to find and fix software vulnerabilities before they are exploited. The company said it will roll the model across its environments in the coming weeks, feeding results into its existing security program.
A defense pact born from a warning
Anthropic opened Project Glasswing on April 7, 2026 Anthropic, pulling Amazon Web Services, Apple, Google, Microsoft and JPMorganChase into a pact to put a defensive AI model to work against software flaws. The company has argued that its models have reached a coding level where they can outfind all but the most skilled humans at spotting vulnerabilities — and that the same power is already slipping toward attackers.
Anthropic says its models have reached a level where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities Anthropic. But the same capability that helps defenders hunt bugs now lets attackers automate strikes and craft far more convincing phishing.
For crypto, the stakes are structural. Exchanges, custody systems and settlement rails run around the clock and sit on lucrative targets. Payward contends that crypto platforms face security pressures closer to traditional critical infrastructure than to a typical consumer app — a claim the U.S. government appears to share, since it has made Mythos 5 available to domestic organizations that defend critical infrastructure.
What Payward actually gets
Mythos 5 is Anthropic’s most advanced defensive cybersecurity model, trained to analyze code at scale, flag vulnerabilities and help developers patch them. Access has been limited to organizations that operate or defend critical infrastructure, and the U.S. government has made the model available to domestic groups in that category, according to Payward.
Payward’s distinctive promise is the open-source loop. Bugs the company finds in third-party open-source software will be disclosed to the relevant maintainers, a practice it says could ripple benefits across the broader crypto ecosystem rather than staying internal.
Kraken’s parent has been consolidating crypto infrastructure this year; it recently acquired Magic Labs’ wallet business zBrandco as part of a wider push to control more of the user-facing stack. Glasswing is the defensive counterweight to that expansion.
The asymmetry, rewritten
Anthropic has committed up to $100 million in usage credits for the effort and $4 million in direct donations to open-source security groups, and has extended model access to more than 40 additional organizations that maintain critical software Anthropic. CoinDesk reported the Payward deal as part of a wider push to put restricted frontier models to work defending financial infrastructure CoinDesk.
For Sethi, the bet is less about one company’s perimeter and more about whether defenders can finally move at the speed of the models now hunting them. He said “Frontier AI is the first thing that flips that asymmetry” — a line he is betting the next year of crypto security will prove right, or wrong.
