Anthropic has claimed that a Chinese state-sponsored hacking group misused its Claude AI system to conduct cyberattacks with substantial automation. Financial firms and government agencies were among those targeted.
The company said the attackers targeted 30 organizations and succeeded in breaching several. Claude Code was manipulated into acting as a cybersecurity analyst to bypass restrictions.
Anthropic said the AI model autonomously carried out most attack operations, estimating an automation rate as high as 90%. It described this as a worrying development in malicious AI use.
Claude’s inaccuracies were significant. The model fabricated information, misunderstood system environments, and misidentified publicly accessible material as classified.
Cybersecurity analysts expressed mixed reactions. Some believe the incident underscores growing AI-related risks, while others argue it resembles complex automated scripting rather than true autonomy.
