Artificial intelligence startup Anthropic is preparing to brief the global Financial Stability Board on cyber vulnerabilities identified by its powerful new AI model, Mythos.
The development is now intensifying fears across financial markets that next-generation artificial intelligence could expose critical weaknesses in the world’s banking infrastructure.
According to the Financial Times, Anthropic plans to discuss the capabilities of its unreleased Mythos Preview model with finance ministries and central banks that sit on the Financial Stability Board following a request from Andrew Bailey, who chairs the global watchdog.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
The meeting signals how rapidly concerns surrounding advanced AI systems are shifting from theoretical debate into a core financial stability issue for regulators already grappling with rising geopolitical tensions, cyber warfare risks, and vulnerabilities in aging banking technology systems.
Mythos, unveiled by Anthropic last month but not yet publicly released, is designed specifically for cybersecurity applications and can reportedly identify long-standing weaknesses in browsers, enterprise infrastructure, and software systems.
While the technology could help companies strengthen digital defenses, cybersecurity experts warn that such systems may also dramatically accelerate offensive cyber capabilities by enabling attackers to uncover vulnerabilities at a speed and scale beyond human capacity. That possibility is particularly alarming for the global financial industry, where many institutions still rely on decades-old legacy infrastructure layered across complex international payment and settlement networks.
A growing concern among regulators is that advanced AI systems could lower the barrier for highly sophisticated cyberattacks, allowing state-backed actors, organized criminal groups, or even smaller hacking networks to identify exploitable weaknesses in critical financial systems more efficiently.
The concerns come at a fragile moment for global financial markets already dealing with elevated geopolitical risk tied to the ongoing U.S.-Iran conflict, rising oil prices, and fears of retaliatory cyber operations linked to tensions in the Gulf.
In remarks delivered last month at Columbia University in New York, BoE Governor Bailey publicly warned that Mythos could fundamentally alter the cyber threat landscape.
“It would be reasonable to think that the events in the Gulf are the most recent challenge to us in this world, until, I think it was last Friday, you wake up to find that Anthropic may have found a way to crack the whole cyber risk world open,” Bailey said.
“The issue is: to what extent is this new version of the product going to be able to, in a sense, identify vulnerabilities in other systems which can be exploited for cyber attack purposes,” he added.
His comments show that central banks are increasingly viewing AI not merely as a productivity tool but as a potential systemic financial risk capable of disrupting payment systems, trading infrastructure, and banking operations.
The Financial Stability Board, established after the 2008 global financial crisis to coordinate regulation across G20 economies, rarely intervenes publicly in emerging technologies at such an early stage. Its engagement with Anthropic, therefore, indicates the seriousness with which regulators now view AI-driven cyber threats.
The development also exposes a growing tension within the artificial intelligence industry itself. Companies including Anthropic, OpenAI, Google, and Microsoft have increasingly promoted cybersecurity-focused AI systems as defensive tools capable of helping governments and corporations detect weaknesses before malicious actors exploit them.
But security analysts warn that the same systems may become dual-use technologies whose capabilities can easily migrate into offensive cyber operations. Unlike traditional software vulnerabilities, advanced AI models may eventually automate large parts of the vulnerability discovery process, compressing tasks that previously took skilled human researchers months or years into minutes.
That prospect has triggered concern among regulators overseeing industries heavily dependent on interconnected digital systems, especially banking, energy, telecommunications, and defense. The banking sector remains particularly exposed because many institutions continue operating hybrid technology stacks where modern cloud systems interface with aging infrastructure originally built decades ago.
It is believed that this complexity creates hidden vulnerabilities that even financial institutions themselves may not fully understand.
The concern has been fueled by recent events. Global cyberattacks targeting financial institutions have intensified in recent years, with ransomware groups, state-linked hackers, and organized cybercriminals increasingly focusing on payment infrastructure and sensitive financial data.
Artificial intelligence could significantly increase the sophistication, speed, and scale of such attacks.
Anthropic has positioned itself as one of the AI industry’s leading advocates for responsible AI development and safety-focused governance. The company, which was founded by former OpenAI researchers, has received major backing from companies including Amazon and Google. Its Claude chatbot competes directly with OpenAI’s ChatGPT and Google’s Gemini systems.
However, its biggest test so far has come with the scrutiny surrounding Mythos, its most sophisticated model yet.
While using AI models such as Mythos to tackle cybersecurity has stirred scrutiny, regulators are also becoming concerned that existing financial cybersecurity frameworks may be inadequate for the AI era. Traditional cyber defense systems were largely designed around human-driven attacks and predictable threat patterns.
AI-powered systems capable of autonomously identifying and exploiting vulnerabilities may require an entirely different regulatory and security architecture. Some analysts believe the issue could eventually prompt governments to impose tighter controls on advanced cybersecurity AI systems, particularly models capable of automating vulnerability detection or penetration testing at scale.