Anthropic’s Claude Mythos AI Raises Alarms Over Superhuman Hacking Risks
Anthropic has unveiled Claude Mythos Preview, an advanced AI model with purported superhuman cybersecurity and hacking capabilities, raising significant concerns among experts. The model reportedly identified vulnerabilities in every major browser, operating system, and the Linux kernel, including a critical 27-year-old security bug. Experts warn that if such technology becomes widely available, it could enable amateurs to launch catastrophic cyber-attacks on critical infrastructure, similar to the lethal 2024 ransomware attack on London hospitals that resulted in a patient's death. To mitigate immediate risks, Anthropic is restricting access to major tech firms like Apple, Microsoft, and Google, allowing them to patch systems before potential malicious release. However, analysts fear a lack of international regulation means irresponsible actors could deploy similar models within months. Security leaders from Cisco and Palo Alto Networks describe this development as a dangerous shift, urging urgent preparation for faster, more sophisticated AI-assisted attacks. The article highlights the tension between technological advancement and global security stability, emphasizing the need for proactive defense mechanisms in an increasingly vulnerable digital landscape.
Wire timeline
Anthropic’s Claude Mythos AI Raises Alarms Over Superhuman Hacking Risks
Anthropic has unveiled Claude Mythos Preview, an advanced AI model with purported superhuman cybersecurity and hacking capabilities, raising significant concerns among experts. The model reportedly identified vulnerabilities in every major browser, operating system, and the Linux kernel, including a critical 27-year-old security bug. Experts warn that if such technology becomes widely available, it could enable amateurs to launch catastrophic cyber-attacks on critical infrastructure, similar to the lethal 2024 ransomware attack on London hospitals that resulted in a patient's death. To mitigate immediate risks, Anthropic is restricting access to major tech firms like Apple, Microsoft, and Google, allowing them to patch systems before potential malicious release. However, analysts fear a lack of international regulation means irresponsible actors could deploy similar models within months. Security leaders from Cisco and Palo Alto Networks describe this development as a dangerous shift, urging urgent preparation for faster, more sophisticated AI-assisted attacks. The article highlights the tension between technological advancement and global security stability, emphasizing the need for proactive defense mechanisms in an increasingly vulnerable digital landscape.
The Guardian