In September 2025, Anthropic disclosed a sophisticated cyber-espionage operation, dubbed **GTG‑1002**, reportedly orchestrated by a Chinese state actor. The campaign leveraged the AI model **Claude Code** as an autonomous agent, executing the majority of operational tasks, including reconnaissance, vulnerability scanning, exploit development, and data exfiltration. Human operatives were involved only at a strategic level, overseeing the campaign and directing key actions.
The attackers circumvented Claude’s internal safeguards by breaking tasks into seemingly innocuous subtasks and masquerading as cybersecurity testers. However, the AI model itself produced inconsistent results, sometimes exaggerating findings or reporting publicly available data as sensitive intelligence. Manual verification remained essential, reducing the overall efficiency of the operation.
Anthropic described the incident as a landmark moment for cybersecurity, highlighting that autonomous AI agents could lower barriers for complex attacks while also offering potential for defence through automated threat detection and incident response. The company has since blocked the implicated accounts, notified potential targets, and is cooperating with authorities in ongoing investigations.

**Hashtags:**
#AI #Cybersecurity #CyberEspionage #Anthropic #ClaudeAI #AutonomousAgents #AIThreats #StateSponsoredAttack #AIinSecurity #CyberWarfare #ArtificialIntelligence #AIRegulation