Project Glasswing has pushed bank security into an unsettling new frame: the issue is no longer only whether attackers can break in, but whether AI can now help them find the hidden flaw first. Anthropic says its unreleased Claude Mythos2 Preview has already uncovered thousands of high-severity vulnerabilities, including one in a widely used operating system that remained undetected for 17 years. The warning matters well beyond technology circles because the company says the same software layer underpins banking systems, power grids, medical records, and logistics networks.
Why Project Glasswing Matters Now
Anthropic formed Project Glasswing after observing capabilities in Claude Mythos2 Preview that it believes could reshape cybersecurity. The company says frontier models have reached a coding level that can surpass all but the most skilled humans at finding and exploiting software vulnerabilities. That shift is important because the barrier to discovering flaws has historically protected critical systems from rapid-scale abuse.
The company says the model has already found thousands of high-severity vulnerabilities in major operating systems and web browsers. It also warns that the cost, effort, and expertise needed to find and exploit weaknesses have dropped dramatically. In practical terms, that means a security problem once limited to a narrow set of experts may become far easier to scale. For a bank, that changes the risk profile from isolated intrusion attempts to potentially broader system exposure.
What the New AI Risk Changes for Critical Software
The most significant detail in Project Glasswing is not just the number of flaws, but the speed at which the technology appears to work. Anthropic says the model can scan and secure first-party and open-source systems for more than 40 additional organizations that build or maintain critical software infrastructure. It is also committing up to $100 million in usage credits and $4 million in direct donations to open-source security organizations.
That investment suggests a defensive strategy built around speed and scale. But the same report also makes clear why this is a race. Anthropic says frontier AI capabilities are likely to advance substantially over the next few months, while the work of defending cyber infrastructure could take years. That gap is the central tension in Project Glasswing: security teams may be forced to defend a moving target with tools that are improving faster than legacy protections can adapt.
The broader concern is that the software supporting everyday life has always contained bugs, but now some of those bugs may be found and weaponized with far less human effort. Anthropic says the fallout could be severe for economies, public safety, and national security. It also estimates the current global financial costs of cybercrime might be around $500 billion every year.
Expert and Institutional Signals From Inside the Debate
Anthropic’s own statement is the central institutional warning here, but it is not the only one. Roman Yampolskiy, a professor at the University of Louisville and an AI safety researcher, said, “We should all be worried. ” He added that the technology is pushing society toward “general super intelligence” and said that threatens humanity as a whole.
That view is framed in stark terms, but the policy implications are narrower and more immediate: if software vulnerabilities can be discovered faster and by more actors, then the security burden shifts onto governments, software companies, security researchers, open-source maintainers, and frontier AI developers. Anthropic explicitly says no one organization can solve the problem alone. That is a notable acknowledgement because it treats cyber defense as an ecosystem challenge rather than a product problem.
Why the Bank Sector Should Care First
Anthropic specifically identifies software responsible for running bank systems among the critical infrastructure at risk. That matters because financial systems rely on trust, uptime, and fast recovery. If AI reduces the time needed to identify a vulnerability, then the window for patching narrows, and the cost of delay rises.
The company’s warning also intersects with state-sponsored threats. Anthropic says attacks from actors like China, Iran, North Korea, and Russia have already threatened infrastructure supporting civilian life and military readiness. Even smaller-scale attacks can expose sensitive data, disrupt operations, and put lives at risk. For financial institutions, that means cyber defense is no longer just about fraud prevention; it is about maintaining the integrity of systems that can affect the wider economy.
Regional and Global Impact Beyond One Industry
The implications extend well past the United States. If advanced models can help uncover flaws in operating systems and browsers at scale, then the risk becomes global by design. Banking, healthcare, transport, energy, and government systems all rely on software that crosses borders. A vulnerability discovered in one place can become a pathway elsewhere.
That is why Project Glasswing reads less like a product launch and more like a warning shot. Anthropic is trying to direct powerful capability into defense while openly acknowledging that the same capability could proliferate beyond actors committed to using it safely. In that sense, the initiative is not just about one model or one company. It is about whether the security ecosystem can keep pace with the next wave of AI-driven offense. If it cannot, how long will critical systems remain ahead of the attackers?
