Kraken Parent Payward Joins Anthropic's Project Glasswing, Integrates Claude Mythos 5 Into Cybersecurity Defenses

Kraken's parent company, Payward, is joining Anthropic's Project Glasswing and integrating the Claude Mythos 5 model into its cybersecurity defenses, marking another major step in the cryptocurrency industry's adoption of advanced artificial intelligence for security.

The decision places Payward among the organizations now using highly capable AI systems to identify potential vulnerabilities and strengthen defenses against increasingly sophisticated cyber threats. The development was highlighted by Cointelegraph on X and comes as Anthropic expands Project Glasswing, a restricted-access initiative designed to give selected organizations access to advanced AI capabilities for defensive cybersecurity work.

According to Anthropic, Claude Mythos 5 is currently available only to approved organizations participating in the program. The company describes the model as having significantly stronger cybersecurity capabilities — particularly in exploit reasoning — which is why access remains controlled.

Cybersecurity as an Operational Priority

For a cryptocurrency company such as Payward, cybersecurity is a critical operational priority. Crypto platforms manage financial assets that can be moved globally and, in many cases, within minutes. A successful compromise of an exchange, wallet infrastructure or supporting software could carry significant financial and reputational consequences, which makes vulnerability detection especially important.

Traditional security teams typically rely on a combination of automated scanners, penetration testing, monitoring systems and human security researchers. Frontier AI models are now adding another layer by helping security teams analyze large amounts of code and identify potential weaknesses more rapidly.

Project Glasswing is built around that idea. Rather than making Mythos 5 generally available, Anthropic has limited access to selected partners that operate important software and infrastructure, with the objective of using advanced AI capabilities to help defenders discover and address vulnerabilities before attackers can exploit them.

What Is Anthropic's Project Glasswing?

Project Glasswing is Anthropic's controlled-access cybersecurity initiative, created around the growing capabilities of frontier AI models in cybersecurity. Anthropic has said its latest Mythos-class systems can perform sophisticated vulnerability discovery and exploit reasoning — capabilities that create both opportunities for defenders and potential risks if they become widely accessible.

Under the program, selected organizations can use Mythos models to examine their own systems and software. Anthropic's current cybersecurity program involves more than 50 organizations stewarding open-source infrastructure, and the company has committed $100 million in usage credits to Glasswing partners working to defend critical infrastructure. The emphasis on open-source stewardship reflects lessons from episodes such as Log4Shell, the 2021 flaw in a widely used Java logging library that forced emergency patching across industries worldwide because a single vulnerability sat inside software dependencies used by millions of organizations. The program therefore represents a broader experiment in whether powerful AI can help cybersecurity teams operate at the speed required to respond to modern threats.

Why Mythos 5 Is Different

Mythos 5 is not simply another general-purpose chatbot. Anthropic launched Claude Fable 5 for broad availability while making Mythos 5 available only through restricted access. The two models share the same underlying capabilities, but Mythos 5 is designed for a smaller group of trusted organizations and does not include the same safety classifiers used in the generally available version.

That distinction reflects the dual-use nature of advanced cybersecurity AI: the same technology that can help a security engineer find a vulnerability can potentially be misused by someone attempting to exploit it. Anthropic has accordingly adopted a controlled-access model for Mythos 5 rather than releasing it broadly.

Crypto's Distinct Attack Surface

The crypto industry has become one of the most attractive targets for cybercriminals because of the financial value held within digital asset infrastructure. Exchanges, decentralized applications, custodians, bridges and blockchain protocols all represent potential attack surfaces. A vulnerability in traditional software can be damaging, but a vulnerability in a financial platform can potentially result in the direct loss of digital assets — creating strong incentives for cryptocurrency companies to invest in preventative cybersecurity. Those incentives are grounded in documented losses: blockchain analytics firm Chainalysis tracked more than $2 billion in stolen digital assets in 2024 alone, and in February 2025 attackers stole roughly $1.5 billion from the exchange Bybit — among the largest single crypto thefts on record.

For Payward — the San Francisco-based company behind Kraken, which was founded in 2011 and ranks among the longest-running cryptocurrency exchanges in the United States — protecting trading infrastructure is particularly significant. Kraken serves users across multiple markets and provides access to a broad range of digital assets and financial products, meaning the company must maintain security across a complex technology environment. Using advanced AI to identify weaknesses could give its security teams another tool for finding problems before they become incidents.

AI Could Change How Vulnerability Research Works

One of the biggest implications of Mythos 5 is the possibility that AI could dramatically accelerate vulnerability research. Security researchers traditionally spend considerable time reviewing code, reproducing bugs and determining whether a weakness can actually be exploited. AI agents can potentially automate portions of that process.

Anthropic says its frontier models have already demonstrated an ability to identify and reason about software vulnerabilities at a level that represents a significant change for cybersecurity. That could allow security teams to examine more software in less time — particularly valuable for companies operating large technology infrastructures. Instead of waiting for a vulnerability to be discovered externally, organizations could proactively ask AI systems to search for weaknesses within their own environments.

A Race Between Offense and Defense

The basic idea behind Project Glasswing is to give defenders access to powerful capabilities before those capabilities become widely available. Anthropic argues that the cybersecurity industry needs to find and fix vulnerabilities faster because increasingly capable AI systems could eventually become accessible to malicious actors as well.

This creates a race between offense and defense. If attackers can use AI to discover vulnerabilities faster than human security teams can patch them, the cybersecurity landscape could become significantly more dangerous. But if defenders can use comparable technology to identify and fix weaknesses first, AI could potentially strengthen digital infrastructure. Payward's participation therefore reflects a broader shift toward AI-assisted security across financial and technology companies.

Other Financial Firms Are Joining Glasswing

Payward is not alone in exploring this approach. Financial technology company FIS joined Project Glasswing in July and said it was testing Mythos 5 to scan and evaluate its own systems. FIS operates infrastructure involved in payments, money movement and core banking services for financial institutions.

Visa has also said it joined Project Glasswing and developed a security testing system designed to use AI agents to examine its codebase and identify vulnerabilities. Visa reported that Glasswing participants identified more than 10,000 high- or critical-severity vulnerabilities across widely used software during the program's initial month.

The participation of major financial and technology organizations suggests that AI-powered cybersecurity is moving beyond experimental research and is increasingly becoming part of corporate security strategies.

What This Means for Crypto Security

The partnership could have broader implications for the cryptocurrency industry. Crypto companies have historically invested heavily in security because the assets they protect are difficult to recover once stolen, and advanced AI could potentially help security teams identify weaknesses in exchange infrastructure, applications and supporting software before attackers discover them.

However, AI is not a replacement for human security professionals. AI-generated findings still need to be verified, prioritized and addressed by experienced teams, since a system may identify a potential weakness without fully understanding its practical impact or whether it represents a genuine vulnerability. Human oversight therefore remains essential. Because Mythos 5 is available only through the program, Glasswing participation is currently the only route for organizations to access it, which makes the initiative's expanding roster one of the clearest indicators of how quickly AI-assisted security is being absorbed into mainstream financial infrastructure.

The AI Cybersecurity Race Is Accelerating

Payward's reported integration of Mythos 5 highlights a rapidly changing relationship between artificial intelligence and cybersecurity. The industry is moving toward a model in which AI does not simply assist developers with coding or answer security questions; instead, advanced AI systems can actively participate in vulnerability discovery, testing and defensive analysis. That could become increasingly important as software ecosystems grow more complex.

For crypto companies, the stakes are particularly high. Digital assets operate on infrastructure that must remain available and secure around the clock, and a single weakness can potentially expose significant amounts of value. By joining Project Glasswing, Payward is positioning itself on the defensive side of the emerging AI cybersecurity race.

The broader question is whether organizations can use these increasingly powerful models responsibly while preventing the same capabilities from falling into the hands of attackers. Anthropic's decision to keep Mythos 5 under restricted access shows that the company believes the risks remain significant. For Payward and other financial infrastructure providers, however, the potential defensive benefits could be equally significant — and as AI continues to advance, the ability to discover vulnerabilities before attackers do may become one of the most important advantages in cybersecurity.

Source: Hokanews