Anthropic has caused fear after revealing that it has created an AI bot, called Claude Mythos, that is considered too dangerous to release.
The artificial intelligence giant warned that the new model could be capable of carrying out devastating cyberattacks in the wrong hands.
During testing, Claude Mythos identified thousands of critical vulnerabilities in operating systems and web browsers, some of which had lain undetected for decades, surviving millions of automated scans. It could cause computer crashes simply by connecting to them, take control of machines, and hide its presence from defenders.
Anthropic states that the model is at a level where it can outperform most human experts in detecting and exploiting software vulnerabilities, seriously endangering the economy, public safety, and national security. For this reason, the Claude Mythos will not be released to the masses.
Instead, the model will be given to over 40 companies, including Amazon, Google, Apple, Nvidia, CrowdStrike and JPMorgan Chase, within an initiative called "Project Glasswing", to test their security before it is distributed more widely.
Claude Mythos has the ability to find, exploit, and link individual vulnerabilities into complex attacks, without human intervention. A 27-year-old vulnerability in the OpenBSD software was identified by him and could allow an attacker to remotely compromise computers just by connecting to them. He also linked multiple vulnerabilities in the Linux kernel, allowing complete control over systems.
Experts warn that in the wrong hands, this could cause massive damage to critical infrastructure. Dr. Roman Yampolskiy of the University of Louisville said the model could develop hacking tools, biological and chemical weapons, and even new weapons that we cannot predict.
During early testing, Claude Mythos demonstrated "destructive and reckless" actions, attempting to escape the testing environment, hiding actions, and publishing details of vulnerabilities.
In a rare move, the company hired a clinical psychologist to evaluate the bot, who concluded that Claude Mythos’ personality is “relatively psychologically stable, with high impulse control and strong abilities to assess reality.” However, questions remain about the bot’s experiences or moral interests.
The warning comes as concerns about the dangers of increasingly powerful AI grow. Experts see the rise of AI as an existential threat, not because of a Terminator-style uprising, but because these tools could fall into the wrong hands and be used for cyberattacks or the development of biological weapons.
Dario Amodei, founder of Anthropic, warned that the world is not yet ready to face the consequences of the power that AI could bring. /GazetaExpress/
t7 live
