Major Crypto Exchange Parent Joins Anthropic’s AI Security Initiative to Hunt Vulnerabilities

Payward Adopts AI-Powered Vulnerability Detection

Payward, the parent organization of major cryptocurrency exchange Kraken, has joined Project Glasswing, Anthropic’s cybersecurity program that grants vetted organizations access to frontier AI models for defensive security work. The company is now deploying Claude Mythos 5, one of Anthropic’s most capable models, to scan its systems for potential vulnerabilities and security weaknesses.

Project Glasswing, according to Anthropic’s documentation, launched in April and expanded in June. Partner organizations participating in the program have collectively uncovered thousands of vulnerabilities rated as high or critical severity. Payward stated that findings identified through this process will route through its existing security division for review and remediation, with patches deployed through standard protocols.

Crypto Industry Mobilizes for Frontier AI Access

Payward’s participation follows a coordinated push by the broader cryptocurrency sector. More than 40 Bitcoin and blockchain companies—including Ark Invest, Coinbase, Block, and BitGo—sent correspondence to major AI laboratories demanding that frontier developers establish trusted-access programs for qualified security researchers. The Bitcoin Policy Institute organized the initiative, arguing that developers protecting open-source financial infrastructure urgently need access to advanced AI capabilities to identify and remediate vulnerabilities before malicious actors exploit them.

Evidence of AI’s cybersecurity potential surfaced when Mozilla disclosed that Claude Mythos identified 271 vulnerabilities in Firefox during evaluation testing. This demonstration of capability at scale strengthened the case for expanded access to these systems among security-focused organizations.

Reframing Defensive Asymmetry

Payward Co-CEO Arjun Sethi characterized the partnership as addressing a fundamental imbalance in cybersecurity dynamics. Attackers need only discover a single exploitable flaw, while defenders must identify and patch every possible vulnerability continuously. Sethi noted that frontier AI represents the first technology capable of processing code with attacker-level precision at machine speed, allowing defenders to locate flaws before exploitation becomes feasible.

Payward also committed to responsible vulnerability disclosure, pledging to report flaws discovered in third-party open-source software to relevant project maintainers. This approach could strengthen security across the open-source ecosystem underpinning cryptocurrency infrastructure globally. As AI-driven security tools become available to major exchanges and financial infrastructure providers, the resilience and trustworthiness of the entire digital asset system stands to improve significantly.

Source: Anthropic (Project Glasswing), via Decrypt. Not financial advice.