Anthropic has opened limited access to its new AI model, Mythos, designed to find critical vulnerabilities and counter cyberthreats. The tool will be available only to select corporations, as its hacking capabilities have been deemed too dangerous for public release.
On communication According to the Wall Street Journal, a preliminary version of the neural network will be distributed to approximately 50 companies and organizations supporting critical IT infrastructure, including Amazon, Apple, Google, and the Linux Foundation. According to Logan Graham, head of Anthropic's Frontier Red Team, the Mythos AI model has demonstrated such advanced capabilities in finding and exploiting software bugs that the developers currently have no plans to release it to the general public.
Cybersecurity experts are increasingly concerned that artificial intelligence is rapidly shrinking the window between the discovery of a vulnerability and the creation of working software to exploit it. Late last year, researchers from Stanford University found that neural networks are almost as effective at finding and exploiting bugs in real networks as human experts. A prime example of this is Anthropic's previous model, Claude Opus 4.6, which earlier this year found more critical vulnerabilities in the Firefox browser in just two weeks than are typically discovered worldwide in two months.
The new Mythos model proved to be approximately 10 times more effective than previous AI systems in calculating the cost of detecting a single bug. Although Anthropic has no plans to release its product publicly, Logan Graham is confident that other neural networks will acquire similar hacking capabilities in the next few years. Therefore, the expert urges the IT community to prepare now for a reality in which the delay between the detection of a software vulnerability and its exploitation in an attack will be reduced to zero.
Source:
Source: 3dnews.ru
