Anthropic: the year's most dangerous AI attacker looked ordinary by technique count

Anthropic analyzed 832 banned accounts and 13,873 observed actions between March 2025 and March 2026. A state-sponsored campaign it disrupted in November 2025 used 30 MITRE ATT&CK techniques yet scored the maximum 100 on Anthropic's own risk scale. The share of medium-risk-or-higher actors rose from 33% to 56% over the year.

• 832 accounts and 13,873 actions mapped across 482 ATT&CK techniques
• November 2025 campaign used 30 techniques but scored 100/100 on risk
• AI-assisted account discovery rose 8.9%, AI-assisted phishing fell 8.6%
• Medium-risk-or-higher actors grew from 33% to 56% in twelve months