•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Anthropic’s newly released AI model, Mythos, is prompting urgent responses from central banks and government agencies worldwide, as the US company tightly controls access to a system viewed as highly powerful and potentially dangerous.
In the two weeks since its emergence, Mythos has triggered a rapid global scramble in the AI era. Anthropic says the model is particularly effective at identifying and exploiting hidden vulnerabilities in software used in banking systems, power grids and other government infrastructure—raising concerns about cyber risk with geopolitical implications for an American provider.
World leaders are working to assess the security threat and determine how to respond. Beyond the United States, Anthropic has shared Mythos only with the United Kingdom.
Public warnings and private requests have followed. The Governor of the Bank of England said the technology could “disrupt the entire cyber risk framework.” The European Central Bank has asked banks to review their defenses, while Canada’s Finance Minister compared the threat to the closure of the Hormuz Strait.
For US rivals including China and Russia, Mythos is also being framed as a reminder of the security consequences of lagging behind in AI development.
AI researchers have long warned that the country leading in the most powerful AI models may gain a geopolitical advantage. In this view, advances are increasingly treated as “weapon-like experiments” rather than routine product launches.
Eduardo Levy Yeyati, former chief economist at the Central Bank of Argentina, said: “As foundational AI models become more important, access becomes a geopolitical issue. This is a policy wake-up call, governments cannot keep turning a blind eye.”
Anthropic says it is restricting access for safety reasons and is currently sharing Mythos with more than 40 organizations tied to critical global infrastructure, including Amazon, Apple and Microsoft. The company said it has no plans to expand access in the near term and will coordinate with the US government and partners on next steps. Anthropic also disclosed it is receiving a large volume of access requests from governments and enterprises worldwide.
In addition, Anthropic forecasts that within 18 months other organizations could release AI models with similar cyberattack capabilities, potentially shortening the time available for defensive systems to prepare.
The US government has taken note of Mythos, despite disagreements with Anthropic over the use of AI in military applications. Anthropic CEO Dario Amodei met White House officials after concerns were raised that the model could cause significant disruption to computer systems.
On April 21, Anthropic said it was investigating reports that some users gained unauthorized access to a Mythos version.
The United Kingdom is the only country outside the US with access to Mythos. The UK AI Security Institute said the model can perform complex cyberattacks that no previous AI had been able to execute.
Within the European Union, Anthropic has engaged multiple times, but access has not been granted due to the lack of a sharing agreement. Germany has not yet accessed the model, though experts described the development as a watershed in the nature of cyber threats.
Public responses in China and Russia have been limited, but AI researchers in China are watching closely. Many Chinese banking, energy and government systems rely on software that Mythos has identified as vulnerable, though they do not yet have access to the technology.
“For China, this could be a second wake-up call after ChatGPT,” said Matt Sheehan, a fellow at the Carnegie Endowment for International Peace.
Some Chinese researchers also worry the country could fall further behind in AI development by missing the advantages of building foundational models.
A spokesperson for the Chinese Embassy in Washington said Beijing does not have full details on Mythos but supported a safe and open cyber space.
The situation is unfolding amid limited international cooperation on AI. The article notes there is currently no global mechanism comparable to a not-proliferation treaty for nuclear weapons to control AI.
Yeyati added that the ability of a company to unilaterally decide who receives access to advanced AI based on non-transparent, non-appealable criteria is concerning.
[Source: Vietstock/FILI]
Premium gym chains are entering a “golden era” that is ending or already in decline, as rising operating costs collide with shifting consumer preferences toward more flexible, community-based ways to exercise. Long-term memberships are shrinking, margins are pressured by higher rents and facility expenses, and competition from smaller, more personalized…
