Anthropic: Approximately 67% of banned accounts used AI to assist in cyberattacks

Anthropic data shows that of the 832 accounts banned for violations in the past year, approximately 67% used AI to assist in cyberattacks, including writing malware. AI has penetrated the entire attack lifecycle, with 6.5% of banned accounts using AI for internal network penetration. The proportion of medium-to-high-risk accounts rose from 33% in the previous six months to 56%. Google discovered the first case of AI being used to develop zero-day vulnerabilities last month. Anthropic is about to launch its AI model Mythos, which has already identified over 10,000 critical vulnerabilities in commonly used software.