アンソロピックの新AI「Mythos」 サイバー攻撃試験で成功率7割
2026-04-13
日本経済新聞
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Anthropic's AI model, Claude Mythos, demonstrated unprecedented autonomous capabilities in discovering and exploiting software vulnerabilities, outperforming human experts in cybersecurity tests. Due to its potential for large-scale cyberattacks, Mythos is not publicly released, prompting heightened defensive measures in sectors like finance and government worldwide.[AI generated]
Why's our monitor labelling this an incident or hazard?
The AI system is explicitly mentioned and was tested for cyberattack capabilities, showing a high success rate. While no actual harm is reported as having occurred, the demonstrated capability and expert warnings about misuse indicate a plausible risk of future harm. Therefore, this event qualifies as an AI Hazard because the AI's use could plausibly lead to incidents involving harm to critical infrastructure or other cyber harms, but no direct harm has yet materialized.[AI generated]