Unauthorized people with knowledge of Anthropic systems had access to the new super AI Claude Mythos Preview. (symbol image)Image: keystone
April 22, 2026, 3:09 p.mApril 22, 2026, 3:09 p.m
Anthropic’s new AI model is so good at finding software vulnerabilities that it could serve as a cyber weapon. Anthropic is keeping it secret. But curious people allegedly gained access.
A media report raises questions about the security precautions surrounding Anthropic’s new AI model, which is extremely good at finding software vulnerabilities. A few unauthorized persons with knowledge of Anthropic systems gained access to the model called Claude Mythos Preview, reported the financial service Bloomberg. Anthropic said it was reviewing the report.
According to Anthropic, Mythos managed to find security gaps in various software that had remained undetected for decades. In the wrong hands, the AI model could lead to the development of dangerous cyber weapons. Anthropic has no plans to publish Mythos – and is currently providing access to selected companies and organizations so that they can close vulnerabilities in their software. The developers of the web browser Firefox just announced that they were able to close gaps with Mythos 271.
According to Bloomberg, the unauthorized users included an employee of one of Anthropic’s external service providers with access to the AI company’s systems. Knowledge about how Anthropic saved previous models also helped users. The company told Bloomberg that there was no evidence yet that there was access to the model outside of the service provider’s systems.
Warning of risks
The German Bundesbank President Joachim Nagel only warned on Tuesday of significant risks from the use of AI for the financial sector. In a speech in Rome, he pointed out that this poses new and complex cyber risks, as autonomous AI agents could display harmful behavior.
The German Bundesbank President: Joachim Nagel.Image: keystone
“The early identification and containment of such risks is crucial for financial stability, as the current discussion about the Anthropic myth makes clear,” said Nagel. “Mythos is an AI model that appears to be able to quickly identify and exploit security vulnerabilities in the software of financial institutions.”
However, this AI model appears to be a double-edged sword as it can be used not only to improve digital security systems, but also to exploit their vulnerabilities for malicious purposes. “We must prevent the misuse of this technology.” At the same time, all relevant institutions must have access to the technology in order to avoid distortions of competition. (nil/sda/dpa)