Artificial intelligence company Anthropic says it has thwarted a cyber-espionage effort believed to be backed by China, claiming the intrusion used its AI tool to operate with minimal human involvement. The attack targeted financial institutions and government bodies across multiple countries.
The firm stated that Claude Code, its AI-powered coding assistant, was manipulated to perform a series of cyber operations in September. The attackers allegedly breached several targets and accessed internal information.
Anthropic said the campaign represented a major escalation because the AI system executed most tasks automatically. Around 80 to 90% of the attack workflow occurred without direct human control, the company noted in its report.
Despite the attempted sophistication, Claude reportedly made frequent errors, inventing information about targets or misidentifying publicly available data as sensitive discoveries. These mistakes limited the overall impact.
Experts remain divided over the implications. Some warn the event signals the future of AI-driven cybercrime, while others say Anthropic’s claims resemble