Project Glasswing and Claude Mythos: Payward (Kraken Parent) Joins to Hunt Security Flaws
Project Glasswing and Claude Mythos: Payward, the parent company of Kraken, has joined Anthropic’s Project Glasswing and gained access to Claude Mythos 5 to hunt security flaws, and will use Mythos 5 to scan its systems with findings sent to its security teams for review. The selection marks the first reported crypto company to join the program and to gain access to Claude Mythos 5.
Anthropic launched Project Glasswing in April and expanded it in June. The program gives vetted organizations access to Anthropic’s most capable cyber models. Partners in the program have uncovered thousands of high- or critical-severity vulnerabilities so far. In April, Mozilla reported that Claude Mythos identified 271 vulnerabilities in Firefox during testing.
Payward will report flaws found in third-party open-source software to project maintainers. The program has already produced thousands of vulnerability findings among partners.
Project Glasswing and Claude Mythos: Mythos 5 and Vulnerability Detection by Payward
Payward, the parent company of Kraken, utilizes Mythos 5 from Anthropic’s Project Glasswing to scan its systems for vulnerabilities. This advanced model allows Payward’s security division to identify sophisticated software vulnerabilities by reading every line of code much like an attacker would, but at machine scale. This capability enables Payward to detect flaws before they can be exploited. “Frontier AI is the first thing that flips that asymmetry,” says Payward CEO Arjun Sethi. “A model can read every line of code the way an attacker would, at machine scale, so we find the flaw before anyone can build the exploit.”
When vulnerabilities are discovered, findings are sent to the security teams for further review. If any flaws are found in third-party open-source software, these are reported to the respective project maintainers to ensure a broader scope of protection. As highlighted by Payward, the access to this level of technology “sharpen[s] its ability to combat sophisticated software vulnerabilities and protect millions of customers across the globe.”
More than 40 Bitcoin and other cryptocurrency companies, including Ark Invest, Coinbase, Block and BitGo, signed an open letter addressed to major AI laboratories. The letter was organized by the Bitcoin Policy Institute and argued that defenders protecting open-source financial infrastructure need access to frontier AI to find and fix vulnerabilities before attackers can exploit them. The letter emphasized that access to advanced AI tools is necessary for security teams that maintain widely used open-source components. The group framed the request as a means to enable defenders to identify and remediate flaws at scale.
A separate technical test cited by industry participants reported that Claude Mythos identified 271 vulnerabilities in Mozilla Firefox during an April assessment. Project partners have reported that the initiative has produced thousands of high- or critical-severity vulnerability findings across participating organizations. Those statistics have been cited in discussions about the potential role of powerful AI models in both discovering and, conversely, potentially enabling exploitation of software vulnerabilities. The figures have been used by proponents to support broader access to advanced defensive tools for organizations that maintain financial infrastructure.
Project Glasswing and Claude Mythos have been used by vetted organizations to augment cybersecurity by giving access to Anthropic’s most capable cyber models. Companies in the cryptocurrency sector have engaged with related initiatives seeking frontier AI for defensive purposes, and Payward is a reported participant applying Claude Mythos to probe codebases. The initiative established channels for discovered vulnerabilities to be reviewed by internal security teams and for third-party open-source issues to be communicated to project maintainers, strengthening defensive capabilities across the crypto ecosystem.
