Claude Mythos Found 271 Zero-Days in Firefox as Bundesbank Demands EU Access

The Firefox team has been working around the clock using frontier AI models to find and fix latent security vulnerabilities in the browser. According to Schneier on Security, an early version of Claude Mythos Preview identified 271 vulnerabilities in Firefox, all of which were fixed in the release of Firefox 150 this week. That follows a previous collaboration where Claude Opus 4.6 found and led to fixes for 22 security‑sensitive bugs in Firefox 148.

The Firefox team described the experience as vertigo‑inducing on Schneier on Security: "For a hardened target, just one such bug would have been red‑alert in 2025, and so many at once makes you stop to wonder whether it's even possible to keep up." Bruce Schneier, writing on Schneier on Security, noted that assuming defenders can patch and push those patches to users quickly, this technology ultimately favors the defenders.

Anthropic launched Claude Mythos Preview earlier in April as part of Project Glasswing, a select group of roughly 50 organizations — including AWS, Apple, Palo Alto Networks, and Nvidia — tasked with testing the AI model. According to InformationWeek, Mythos is being used by Anthropic and Project Glasswing to identify and exploit zero‑day vulnerabilities in open source codebases.

Mythos's capabilities go well beyond traditional static analysis. Anthropic's own testing found that Mythos is "capable of identifying and then exploiting zero‑day vulnerabilities in every major operating system and every major web browser when directed by a user to do so." The tests even identified some vulnerabilities that are over 20 years old. According to Gartner, less than 1% of potential vulnerabilities uncovered by Mythos have been fully patched by their maintainers.

Germany's Bundesbank is pushing for European banks to be given access to Anthropic's Mythos model for cybersecurity defense. According to Reuters, European banks need access to Mythos if they are to shield themselves from increasingly sophisticated cyber attacks. The Bundesbank's position marks a significant shift: rather than treating Mythos as a threat to be contained, one of Europe's most powerful central banks sees it as a defensive necessity.

This demand comes amid Anthropic's decision to withhold Mythos from general release over cybersecurity concerns — a move that Capital Brief describes as prompting debate over whether the model represents "a wake‑up call on AI or marketing hype." The Bundesbank's stance effectively argues that restricting access hurts defenders more than attackers, since sophisticated adversaries will eventually develop similar capabilities regardless.

The emergence of Mythos is forcing a fundamental rethink of vulnerability management. Kara Sprague, CEO of HackerOne, told ¹ that the gap between when a vulnerability is discovered and when an adversary has a working exploit "has collapsed." For years, security teams relied on that gap to patch systems before attacks materialized — AI has eliminated that buffer.

Mythos can also chain together lower‑severity findings into complex exploits, Sprague noted. Dennis Xu, an analyst at Gartner, described the capability to generate working exploit code to breach enterprise systems as "previously unheard of" from frontier language models. The implication for builders: every piece of software you ship needs to assume that AI‑powered adversaries can find and exploit vulnerabilities faster than any human review cycle.

The dual‑use nature of Mythos is at the heart of the current debate. Anthropic's position is that releasing Mythos to a select group first gives cybersecurity defenders a head start. "Once the security landscape has reached a new equilibrium, we believe that powerful language models will benefit defenders more than attackers, increasing the overall security of the software ecosystem," Anthropic said, as reported by InformationWeek.

Not everyone is convinced. Bad actors are already using AI to "develop more sophisticated AI‑malware and accelerated adaptive attack campaigns," according to a report by research firm Omdia cited in the same article. The Bundesbank's demand for EU access reflects a pragmatic middle ground: since AI vulnerability discovery is inevitable, the priority should be ensuring that legitimate defenders — especially critical infrastructure like banks — get access first.

Marketplace reports that in the cat‑and‑mouse game of cybersecurity, advanced AI like Mythos could help both sides — the question is which side gets more from it.

The Firefox experience offers a blueprint for how AI vulnerability discovery will reshape software development. The Firefox team noted that organizations may need to, per Schneier on Security, "reprioritize everything else to bring relentless and single‑minded focus" to security patching. For builders, that means:

• Assume AI will find your bugs. If Mythos found 271 vulnerabilities in a hardened browser, it will find them in your application too. Static analysis and manual code review are no longer sufficient.
• Patch velocity matters more than ever. The collapsed window between discovery and exploitation means your deployment pipeline needs to ship security fixes in hours, not weeks.
• Advocate for defender access. If you work in critical infrastructure, the Bundesbank's demand for Mythos access is a model worth following. AI security tools should not be restricted to a handful of big tech companies.