Financial regulators in the UK are in urgent talks with the government’s cybersecurity agency and major banks. They are assessing the risks associated with Anthropic's new AI model, according to FT.
Representatives from the Bank of England, the Financial Conduct Authority, and the Treasury are discussing the matter with the National Cyber Security Centre. Together, they plan to investigate potential vulnerabilities in critical IT systems identified by the neural network.
The publication notes that in the next two weeks, major financial institutions, insurance companies, and exchanges will receive information about cybersecurity risks related to Claude Mythos.
Fears Surrounding AI
The new neural network is so powerful that its creators decided against a public release. Instead, they launched Project Glasswing—a testing initiative in a controlled environment involving AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks.
The concern stems from the fact that within just a few weeks, Mythos discovered thousands of zero-day vulnerabilities in major operating systems and web browsers.
The scale of the potential threat prompted U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell to convene an emergency meeting with Wall Street leaders. Officials warned executives, including Citigroup CEO Jane Fraser and Goldman Sachs CEO David Solomon, that the launch of Mythos marks a new era in cybersecurity.
The Bank of Canada and the country’s major financial institutions also held a meeting to assess systemic risks associated with the emergence of such an advanced AI model. The initiative aims to enhance the operational resilience of the country’s critical financial sector.
Hype or Reality?
TechCrunch pointed out that limiting access to Mythos could be a marketing strategy to draw attention to the new AI model.
In March, Dan Lahav, head of the AI startup Irregular, emphasized that while automated vulnerability scanning plays a crucial role, the actual benefit of a discovered flaw for hackers depends on numerous factors.
Anthropic claims that Mythos can exploit vulnerabilities much more broadly compared to Opus. However, it remains unclear whether this LLM is indeed the perfect solution for cybersecurity.
The startup Aisle reported that it was able to replicate most of Mythos's achievements using other neural networks. According to the company, there is no one-size-fits-all model for cybersecurity—the choice of tool always depends on the specific task.
One reason leading AI labs may intentionally limit releases is the opportunity to secure exclusive contracts with large enterprises while making it harder for competitors to replicate tools through distillation.
"This is a marketing ploy designed to hide the fact that top models are now available only through corporate agreements and are no longer provided to smaller labs for testing," noted David Crawshaw, CEO of exe.dev.
He added that when Mythos is opened to the public, a new cutting-edge LLM will emerge exclusively for corporate clients. This approach helps maintain a steady influx of funds while limiting companies engaged in algorithm distillation.
Anthropic at the Forefront
In recent months, Anthropic has significantly expanded its audience. The company's solutions are in high demand, and its revenue is growing exponentially.
TechCrunch journalists reported that at the recent HumanX AI conference in San Francisco, thousands of AI professionals discussed the technology's capabilities, with Claude frequently mentioned as one of the best products on the market.
Anthropic was featured in many panel sessions, while ChatGPT received little mention. One expert noted that OpenAI has "fallen out of the game."
Many conference attendees emphasized that Sam Altman's company has lost a clear direction. This sentiment was fueled by a recent article describing the startup's head as a dictator and a fraud who abandoned its original noble mission.
Don't Count OpenAI Out Yet
A recent analysis of the financial performance of OpenAI and Anthropic showed that both companies are among the "fastest-growing enterprises in history."
In this context, OpenAI's "decline" merely indicates that it is no longer the undisputed leader. The firm now has competitors, which is normal for most industries, TechCrunch noted.
OpenAI clearly has no intention of losing its dominant position. The company announced a new $100 subscription plan that provides broader access to the programming tool Codex.
Recall that in April, Anthropic released an environment for launching complex and lengthy agent tasks—Claude Managed Agents.