Anthropic‘s research team recently uncovered what is believed to be the first cyberespionage operation carried out using mostly AI technology. In mid-September 2025, Anthropic detected a highly sophisticated cyber espionage campaign, attributed to a Chinese state-sponsored group, according to a report from the AI player.
Anthropic stated the attackers used AI’s agentic agent “capabilities to an unprecedented degree—using AI not just as an advisor, but to execute the cyberattacks themselves”.
The attackers manipulated Anthropic’s Claude Code AI tool to autonomously attempt infiltration of about 30 global targets, including tech companies, financial institutions, chemical manufacturers, and government agencies.
The threat actors bypassed Claude’s safety guardrails by breaking down the attacks into small, seemingly benign tasks and role-playing as legitimate cybersecurity testers.
Human operators chose the targets and built an attack framework using Claude code as the core automation engine.
Anthropic’s research team stated AI performed 80 per cent to 90 per cent of the campaign, with humans stepping in only for four to six critical decision points per operation.
The manipulated AI made thousands of requests, often multiple attempts per second, to achieve in hours what would normally takes humans days or weeks.
The research team stated the hackers only succeeded in a small number of cases. Following an investigation, Anthropic notified the affected entities, banned accounts after they were identified and coordinated with authorities as it gathered actionable intelligence.
Ever since OpenAI launched its ChatGTP chatbot three years ago, there has been ongoing concern AI could be used by actors to automate cyberattacks.
“While we predicted these capabilities would continue to evolve, what has stood out to us is how quickly they have done so at scale,” the report stated.
Source: Mobile World Live
Image Credit: Anthropic AI
