en.Wedoany.com Reported - Artificial intelligence startup Anthropic has agreed to brief members of the Financial Stability Board (FSB) on cyber defense vulnerabilities within the global financial system identified by its AI model, Mythos. The FSB is currently drafting a report on sound practices for the application of AI in the financial system, scheduled for release next month for public consultation.
Mythos is Anthropic's frontier AI model, Claude Mythos Preview, released in April this year. Due to its excessively powerful real-world capabilities, the company decided not to make its model capabilities available to the public, providing controlled access only to approximately 40 critical infrastructure partners through Project Glasswing. In a blog post, Anthropic stated that the model has discovered thousands of significant vulnerabilities in operating systems, web browsers, and other widely used software.
The Financial Stability Board is a global financial regulatory body established by the Group of Twenty (G20) in June 2009, headquartered in Basel, Switzerland. Its members include central banks, finance ministries from G20 nations, and international organizations; the People's Bank of China and the Ministry of Finance of China are both member institutions. The FSB coordinates financial regulatory rules for G20 economies, with Bank of England Governor Andrew Bailey serving as its Chair.
Bank of England Governor Bailey had previously publicly warned that Mythos could pose a significant risk to the entire cybersecurity landscape. Speaking at an event at Columbia University, he indicated that Mythos might have found a way to crack the entire world of cybersecurity risk, with the key question being the extent to which the model can identify vulnerabilities in other systems that could be exploited for cyberattack purposes. Reports indicate that this briefing arrangement was made at Bailey's request to Anthropic.
In February this year, Anthropic had already reported a batch of vulnerabilities discovered in Firefox through the Claude model to Mozilla. Subsequently, Mozilla built an agent-based vulnerability scanning framework on its own fuzzing infrastructure, importing models like Claude Mythos Preview for large-scale scanning. In April 2026, Firefox patched a total of 423 security vulnerabilities, of which 271 were discovered by Mythos, with 180 reaching high-severity levels. Compared to the same period in 2025, when only 31 vulnerabilities were patched, this represents a 13-fold increase. Beyond Firefox, Anthropic's internal red team testing reports show that Mythos achieved an 83% success rate in generating usable exploits on its first attempt, and successfully completed 6 out of 10 attempts in a 32-step simulated corporate intranet penetration test conducted by the UK AI Safety Institute.
Compared to traditional AI tools, Mythos differs in its ability to continuously discover and exploit zero-day vulnerabilities without human guidance. Validation from Anthropic's collaboration with Mozilla shows that vulnerability reports generated by the AI model have almost no false positives, though the Mozilla team also pointed out a real-world limitation: AI currently cannot truly fix vulnerabilities automatically, and the repair code it generates is usually not directly deployable, still requiring human engineer intervention.
Anthropic's briefing to the FSB comes against the backdrop that Mythos may pose a systemic cybersecurity threat to the global financial system. The financial industry is highly dependent on legacy technology systems, and its defensive capabilities will be severely strained when faced with AI-driven bulk vulnerability discovery and automated attacks. Experts warn that Mythos could make advanced cyberattack capabilities readily accessible. At the FSB level, this is no longer just a security issue for individual financial institutions, but a systemic risk to the global financial system. The FSB's 2026 work program has already listed digital innovation and artificial intelligence as top priorities. Previous reports published by the FSB have outlined the potential impact of AI applications in finance on financial stability. The forthcoming report on sound AI practices is expected to further propose a regulatory framework, and Anthropic's briefing to the committee provides regulators with a direct opportunity to understand the capabilities of frontier AI in cyber offense and defense.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
