In an extraordinary move reflecting growing concern over escalating digital risks, US Treasury Secretary Scott Bisnett recently convened an urgent meeting with Federal Reserve Chairman Jerome Powell and top executives from major US banks, warning them of increasing cyber threats driven by a new artificial intelligence model developed by Anthropic.
During this closed-door meeting in Washington, Bisnett delivered a firm message to the leaders of major financial institutions, including Bank of America, Citigroup, and Wells Fargo, warning that allowing the new Cloud Mythos Preview program to run on the banks’ internal computer systems could pose a serious risk to sensitive customer data.
According to a report by The New York Times, reviewed by Sky News Arabia, the bank executives were informed at the meeting that the Cloud Mythos Preview program is highly effective at identifying security vulnerabilities within banking systems, to the point where it raises concerns about the possibility of hackers or malicious third parties accessing and exploiting this information.
According to Kevin A. Hassett, director of the U.S. National Economic Council, stated that the U.S. administration is taking all possible steps to ensure everyone’s safety from the potential risks arising from the Cloud Mythos Preview program. This includes securing Anthropic’s agreement to postpone the public release of its new software until officials can thoroughly review all aspects of it.
Anthropic has already announced that the developments in its Cloud Mythos Preview model are too critical and sensitive to be safely released to the public at this time. The company indicated that Cloud Mythos will be restricted to a consortium of 40 companies called “Project Glasswing.”
This consortium includes a range of companies and institutions operating in various sectors, including at least one bank: JPMorgan Chase, the largest bank in the United States, which announced it will use the software to evaluate next-generation AI tools for defense-grade information security in critical infrastructure.
What is Cloud Mythos?
Anthropic has released little information about its new model, codenamed “Capipara” during its development.
However, the company has acknowledged that Cloud Mythos represents a “quantum leap” in AI capabilities, with improved performance in areas such as programming and cybersecurity research.
According to Anthropic, Cloud Mythos is already capable of conducting independent security research, including finding and exploiting unknown vulnerabilities in critical software—vulnerabilities unknown even to the software’s developers. The company claims that the new model has already identified thousands of bugs and vulnerabilities in popular software systems, including every major operating system and browser.
Anthropic also announced that Cloud Mythos discovered a 27-year-old vulnerability in the OpenBSD operating system, which powers many internet routers and firewalls, as well as another long-standing issue in a popular video program that had been scanned five million times by automated testing tools without finding any flaw. Logan Graham, who heads the team testing the dangerous capabilities of new AI models at Anthropic, says that Claude Mythos represents a starting point for what he believes will be a historic turning point in the AI industry.
Who can access Claude Mythos?
Anthropic says that Cloud Mythos possesses exceptional capabilities that preclude its public release.
Instead, Anthropic announced that it will make the new model available to a consortium of more than 40 technology companies, including Apple, Amazon, Microsoft, Cisco, Broadcom, and Linux.
Anthropic clarified that it does not plan to deploy its new technology more widely. Jared Kaplan, the company’s chief scientific officer, stated that the goal is to raise awareness and give well-intentioned actors a competitive advantage in securing infrastructure, open-source software, and proprietary software.
A precedent is being set. Anthropic’s decision to postpone the public release of Cloud Mythos for fear of misuse is not without precedent. In 2019, OpenAI announced the development of the GPT-2 model but did not immediately release the full version due to its script-generating capabilities, which could be used to automate the mass production of propaganda or disinformation. OpenAI later released the model after conducting additional security tests. Anthropic’s Revenues More Than Triple
Anthropic occupies a unique position in today’s artificial intelligence landscape. It’s racing to build powerful AI systems and earning billions of dollars from selling access to these systems, while simultaneously highlighting the risks its technology poses. The Pentagon designated the company a supply chain risk this year due to its demands for certain restrictions on the use of its technology. A federal judge later suspended this designation.
Anthropic recently announced that its projected annual revenue has more than tripled in 2026, exceeding $30 billion from $9 billion. This growth is largely attributed to the popularity of Anthropic’s Cloud software as a programming tool.
A Positive Shift
Hilda Malouf Melki, an Oxford-certified AI consultant, told Sky News Arabia that what we are witnessing today regarding the Cloud Mythos AI model should not be viewed with fear, but rather as a profound positive shift. The real power this model has demonstrated in uncovering a 27-year-old security vulnerability, or detecting technical problems that millions of automated tests failed to identify, means the world now possesses an artificial intelligence capable of cleaning up the digital space from vulnerabilities that have accumulated over decades. This represents a qualitative leap in the balance of cyber power.
Strategic Postponement
Maalouf Maliki believes that postponing the public release of the Cloud Mythos model does not reflect a fear of artificial intelligence itself, but rather demonstrates a high level of maturity in managing these technologies. This strategic move aims to give well-intentioned entities a competitive advantage in terms of timing, as the immense power of Cloud Mythos will be used first to fortify global infrastructure, banks, and operating systems before it becomes publicly available. She points out that this approach reflects significant maturity in the technology sector, as companies today are not only pursuing financial profit, but also prioritizing national and financial security.
Protecting the Financial Nerve
According to Maalouf Maliki, the warning issued to banks should not be interpreted as a sign of danger. The primary reason behind this warning is the sensitivity of the banking sector, which deals with enormous amounts of financial and personal data. This data is a tempting target for any malicious actor. Therefore, introducing such a powerful tool into banking systems without strict controls and thorough testing could open the door for these actors to exploit the situation. It’s worth noting that any breach of this data not only threatens financial institutions but also directly impacts the stability of the global financial system as a whole. Hence, this warning serves as a proactive regulatory step to ensure that these technologies are integrated according to the highest security and oversight standards, thus preventing any negative repercussions.
Robust Financial Systems
Maalouf Melki emphasizes that Cloud Mythos will give banks a valuable opportunity to understand and test their capabilities in secure environments, contributing to the building of more robust systems in the long term. She notes that the meeting between the US Treasury Secretary and the Federal Reserve Chair with banking leaders to discuss this technology demonstrates that the government and the private sector are working together to ensure that the levers of power remain in safe hands. With this approach continuing, artificial intelligence will not be a threat, but rather the first line of defense for protecting individuals and institutions in the digital age.
A Historic Turning Point
Technology expert Mohammed Aitani told Sky News Arabia’s “Economy” website that the emergence of models like Cloud Mythos represents a historic turning point in cybersecurity doctrine. After decades of relying entirely on human beings and an elite group of bug bounty hunters, who alone possessed the keys to discovering deep vulnerabilities, we now have artificial intelligence with comprehensive analytical capabilities. It operates in the blink of an eye, searching for errors and deducing complex weakness patterns that might escape even the most intelligent human minds. This explains Cloud Mythos’s discovery of vulnerabilities that had remained dormant for a quarter of a century.
Greater Security
Aitani points out that the government and banking sector’s focus on the issue of postponement and testing does not reflect a fear of technology, but rather a progressive understanding that the true value of these models lies not only in their rapid deployment. An additional day of testing Cloud Mythos within closed environments means a reduction in the likelihood of risks on a broader scale later on. He notes that the involvement of entities such as central banks and the Federal Reserve in the discussion reflects the transformation of artificial intelligence from a purely technical matter to one of financial and security stability simultaneously—a situation that was not previously the case. Major institutions no longer treat cybersecurity as a traditional layer of protection, but as a dynamic system that changes as attack and defense tools evolve.
Continuous Monitoring
According to Aitani, one of the most important transformations that these models will bring about
