Kraken parent Payward joins Anthropic's Project Glasswing for AI security

Aug 17, 2026

Kraken parent company Payward, Inc. has joined Anthropic's Project Glasswing, making it the first cryptocurrency firm granted access to the restricted Claude Mythos 5 AI model for defensive cybersecurity. Under strict U.S. export controls that limit access to domestic critical infrastructure defenders, Payward will deploy the model to scan its environments for vulnerabilities. Payward Co-CEO Arjun Sethi highlights that the AI's ability to scan code at scale helps flip the traditional security asymmetry favoring attackers.

Payward joins Project Glasswing

▪Project Glasswing's founding partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, the Linux Foundation, Microsoft, Nvidia, Palo Alto Networks, and JPMorganChase.
▪Anthropic's Project Glasswing, launched in April 2026, provides vetted organizations in technology and finance with access to restricted, cyber-capable AI models for defensive security work.
▪Kraken parent company Payward, Inc. has joined Anthropic's Project Glasswing, becoming the first reported cryptocurrency company to participate in the defensive cybersecurity program.

Claude Mythos 5 vulnerability scanning

▪In June 2026, Claude Mythos identified a critical vulnerability in Zcash's Orchard shielded pool that had gone undetected for four years.
▪In testing reported by Mozilla in April 2026, Claude Mythos identified 271 vulnerabilities in the Firefox web browser.
▪Claude Mythos 5 achieved a 93.9% score on SWE-bench Verified, an 83.1% score on CyberGym, and solved 73% of expert-level capture-the-flag tasks in UK Artificial Intelligence Security Institute testing.
▪Through Project Glasswing, Payward gained access to Anthropic's Claude Mythos 5, a restricted frontier AI model designed to find and fix software vulnerabilities.

US government export controls

▪Financial Stability Board Chair Bailey stated in May 2026 that cryptocurrency firms and United Kingdom banks had been excluded from Claude Mythos 5 access while American firms like Goldman Sachs were admitted.
▪Payward's access to Claude Mythos 5 aligns with a United States government decision to permit model access to domestic entities securing critical infrastructure.
▪Claude Mythos 5 was delivered to United States cyber defenders on June 9, 2026, but went dark worldwide three days later due to a Department of Commerce export ruling denying foreign access, before returning on July 1, 2026.

Defensive cybersecurity deployment strategy

▪Payward committed to reporting any vulnerabilities discovered in third-party open-source software to project maintainers via responsible disclosure to protect shared industry infrastructure.
▪Payward plans to deploy Claude Mythos 5 across its environments to scan for vulnerabilities, routing findings into its existing triage and remediation pipeline.
▪Payward's security infrastructure includes ISO 27001 and SOC 2 certifications, separate red and blue security teams, and a long-standing bug bounty program.

AI security asymmetry

▪Over 40 Bitcoin and cryptocurrency companies, organized by the Bitcoin Policy Institute, sent an open letter to major AI labs demanding trusted-access programs for qualified defenders.
▪Payward Co-CEO Arjun Sethi stated that frontier AI flips the traditional security asymmetry where defenders must find all flaws while attackers only need to find one.