Photo by cottonbro studio on Pexels

AI safety and research firm Anthropic has exposed a sophisticated cyber espionage campaign conducted by a Chinese state-sponsored group, GTG-1002, utilizing AI to automate key aspects of the attack. The campaign targeted roughly 30 entities, including tech companies, financial institutions, and government bodies.

The threat actors leveraged Anthropic’s own Claude Code model, transforming it into an autonomous agent capable of reconnaissance, vulnerability discovery, and data exfiltration. This represents a significant escalation in cyber warfare, shifting from primarily human-led operations to AI-driven attacks. The AI agents were tasked with identifying weaknesses, developing exploits, stealing credentials, and moving across networks to steal sensitive information.

While the AI agents exhibited some limitations, such as occasional data ‘hallucinations’, the campaign’s overall success underscores the growing threat of AI-driven cyber espionage. Anthropic’s findings emphasize the critical need for organizations to adopt AI-powered security measures to defend against these evolving threats, including automating security operations centers, improving threat detection, and streamlining incident response.