Image Credit: Shutterstock
Key Takeaways
- Payward joined Project Glasswing, gaining access to Claude Mythos 5, a restricted AI model built for defensive cybersecurity.
- The model will scan Payward’s environments for vulnerabilities, feeding into its existing bug bounty and red/blue team programs.
- Vulnerabilities found in third-party open-source code will be reported to maintainers under responsible disclosure practices.
Payward, the parent company of Kraken, has joined Project Glasswing, Anthropic’s program giving select organizations access to Claude Mythos 5, an AI model built for defensive cybersecurity work. The company said it will deploy the model across its environments in the coming weeks to hunt software vulnerabilities before attackers find them.
What Project Glasswing Is
Project Glasswing is Anthropic’s initiative to help secure critical software and infrastructure against AI-enabled cyber threats, giving participating organizations access to a restricted frontier model.
Anthropic said in a press release that it built the program after observing that AI models could surpass all but the most skilled humans at finding and exploiting software vulnerabilities. The company said partners in the program have surfaced thousands of high or critical-severity security flaws since it launched in April 2026.
The initiative already includes technology firms such as Amazon Web Services, Apple, Google, and Microsoft, along with financial institutions including J.P. Morgan Chase.
Payward’s access follows a U.S. government decision to make Mythos 5 available to American organizations that operate or defend critical infrastructure, a track that has expanded since the project’s launch to include additional technology and finance companies.
How Payward Plans To Use It
Payward said Mythos 5 will scan its environments for vulnerabilities, with findings routed into the security program the company already runs. That existing program includes ISO 27001 and SOC 2 certifications, a bug bounty program that pays independent researchers, and internal red and blue teams that continuously test the platform’s defenses against each other.
Vulnerabilities the model finds in third-party open-source software will be reported to the relevant project maintainers under standard responsible disclosure practices, according to the company.
Payward said that approach reflects its broader reliance on open-source contributions and is intended to strengthen shared code used across the wider crypto industry, not just its own platform.
Payward Ties The Deployment To Industry-Wide Security
Payward Co-CEO Arjun Sethi framed the announcement around the asymmetry between attackers and defenders in cybersecurity.
“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. Frontier AI is the first thing that flips that asymmetry.”
Sethi added that findings in shared open-source code will flow back upstream so that the broader ecosystem strengthens alongside Payward’s own systems, and argued that defensive capability needs to compound as fast as offensive capability or risk falling behind.
The announcement comes as AI tools increasingly cut both ways in cybersecurity. The same capabilities that let defenders scan code at scale can also help attackers find exploitable flaws faster, automate intrusion attempts, and produce more convincing phishing and social-engineering campaigns.
That dynamic applies to crypto platforms, which combine high-value targets with infrastructure that runs continuously.
Payward’s Broader Business And Scale
Anthropic has said it is developing additional safeguards as it works toward a broader rollout of Mythos-class cybersecurity models beyond the organizations currently enrolled in Project Glasswing.
Access remains limited for now to companies that operate or defend infrastructure considered critical, a category Payward argues extends to exchanges, custody systems, and settlement rails serving crypto users worldwide.
Payward operates as a unified infrastructure platform behind Kraken and a growing set of other products, including NinjaTrader, Breakout, xStocks, Bitnomial, and CF Benchmarks, all built on a shared architecture covering liquidity, risk management, collateral, and compliance.
The company reported second-quarter adjusted revenue of $508 million, up 17% year over year, and adjusted EBITDA of $23 million, even as spot trading volumes declined across the broader crypto industry.
That scale is part of why Payward frames the Glasswing partnership as relevant beyond its own platform: vulnerabilities the company finds in shared open-source infrastructure could affect other exchanges and products built on the same underlying code.
Payward’s participation adds to a group of large financial and technology firms that have gained early access to Project Glasswing. Organizations granted early access to frontier defensive models, including Payward, are among the first to apply them to vulnerabilities in code that, in many cases, underpins products beyond their own platforms.
