Anthropic restricts the use of the new model to giant companies only

Among the companies that will get access to the model, called Mythos (or “Claude Mythos Preview”), are Amazon, Apple, Cisco, Google, JPMorgan Chase and Microsoft. The model will be used to detect software weaknesses and to test the effectiveness of hacking techniques on existing systems.

A powerful model – but dangerous for advertising

According to the company, the model is not ready for launch to the general public due to the risks of misuse by cybercriminals and spy agencies – a concern that also worries officials in Washington and Silicon Valley. Experts point out that the ability of AI agents to scan for weaknesses at a rate and scope far exceeding human capabilities constitutes a dramatic change in the cyber field.

Anthropic informed senior officials in the US government about the model’s offensive and defensive capabilities, and offered assistance in testing and evaluating the technology. Along with the technology giants, companies like Broadcom, Nvidia, Linux Foundation as well as cyber companies like CrowdStrike and Palo Alto Networks will get access to the system.

The main goal is to find critical weaknesses in browsers and common operating systems even before they are revealed to the public, thereby reducing the attackers’ advantage.

Thousands of weaknesses in a short time

According to the company, the model has already managed to locate “thousands” of previously unknown software weaknesses – at a rate much faster than that of human security researchers, although this figure has not yet been independently verified. Weaknesses of this type are considered a valuable asset for hackers and intelligence agencies, as they allow covert penetration of systems.

According to company executives, if such models continue to improve – as expected – the world will have to quickly prepare for a new reality. A recent leak even claimed that the model is “far ahead” of other models in the cyber field, implying an approaching wave of systems capable of exploiting weaknesses at a rate that will overtake defense capabilities.

Alongside this, experts estimate that some of the concerns are exaggerated, but warn that without a change in trend – the gap between attackers and defenders in the age of artificial intelligence may widen even further.