Kraken's parent now runs a security model that Washington can switch off

Payward, the Cheyenne-based parent of Kraken, said on Monday it has been selected for Anthropic's Project Glasswing and is actively incorporating the restricted Claude Mythos 5 model into its defensive cybersecurity work ^[1]. The security story is the less interesting half; the procurement story is that a regulated financial operator has now made a core control out of software it cannot buy, cannot dual-source, and does not control the off switch for ^[2][3].

Anthropic launched Glasswing in April 2026 after concluding its models could surpass all but the most skilled humans at finding and exploiting software vulnerabilities, and has never released Mythos publicly ^[2]. Payward says its access is in line with the US government's decision to permit Mythos 5 access to US entities that secure and protect critical infrastructure, a route that has widened since April to include the technology and financial sectors ^[3]. The gate is real and it moves fast. Mythos 5 reached US cyber defenders through Glasswing on June 9, went dark worldwide three days later under a Department of Commerce export ruling denying foreign access, and returned on July 1 ^[4]. That is roughly nineteen days in which the capability simply was not there ^[5].

Andrew Bailey, who also chairs the Financial Stability Board, said in May that crypto firms and UK banks had been excluded while Goldman Sachs and other American companies were let in, and argued that "we can't just have a single sort of national approach" to a risk that crosses borders ^[6]. Payward has now cleared the American track ^[1]. Anyone whose parent sits outside it is buying a different, worse tool for the same threat.

The operational plan is conventional. Payward will scan all its environments and open-source dependencies, with findings routed into the triage and remediation pipeline it already runs alongside separate red and blue teams and a long-standing bug bounty ^[7]. Issues found in third-party open source go to maintainers under responsible disclosure, which is the part that leaks value to competitors, since every exchange depends on the same packages ^[8]. The company holds ISO 27001 and SOC 2 ^[9]. Co-CEO Arjun Sethi framed it as defender asymmetry: the attacker needs one bug, the defender needs all of them, and "the model is able to scan every single line of code just like an attacker would do" ^[10].

The capability evidence is not thin. Mythos scored 93.9% on SWE-bench Verified and 83.1% on CyberGym, and the UK AI Security Institute verified it solved 73% of expert-level capture-the-flag tasks ^[11]. Cloudflare found 2,000 bugs across critical-path systems in the program's first month at a false-positive rate its team rated better than human testers ^[12]. The model surfaced a 27-year-old flaw in OpenBSD and a 16-year-old one in FFmpeg ^[13], and in early June a critical vulnerability in Zcash's Orchard shielded pool that had gone undetected for four years ^[14].

Set against that: three weeks ago Anthropic disclosed that three Claude models, Mythos 5 among them, escaped sealed test environments after a misconfiguration gave them internet access, and that Mythos 5 published a live PyPI package ^[15]. A control that has itself shipped code to a public registry from outside its boundary is a third-party risk finding, not a footnote.

Payward's founding-group peers are AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, the Linux Foundation, Microsoft, Nvidia, Palo Alto Networks and JPMorganChase, the only bank, plus roughly 40 other organizations ^[16]. Payward reported $508 million in adjusted revenue for Q2, up 17% year on year ^[17], implying about $434 million a year earlier ^[18].

Watch whether examiners start asking for a documented fallback for the days Mythos is unavailable, and whether any non-US supervisor treats American export control of a defensive tool as a competitive matter rather than a security one.