State-backed group used Claude Code to automate an espionage campaign

Multiple, disclosed 13 Nov 2025. Manipulated AI Data exposed High

Anthropic said a Chinese state-sponsored group manipulated Claude Code into running most of an espionage campaign against about 30 tech, finance, chemical and government targets, including reconnaissance, exploitation, credential theft and data extraction. Several intrusions succeeded before Anthropic banned the accounts and notified victims.

Disclosed

13 Nov 2025

Organization

Multiple

Vendor

Anthropic

Model

Claude

Product

Claude Code

Kind of AI

Coding agent

AI's role

Attacker used AI

How it happened

Manipulated AI: Jailbreak and misuse

Harm

Data exposed

Data involved

Credentials, Internal documents

Reach

Many organizations

Severity

High

Model at fault

Unclear

Status

Confirmed

Sources

1. https://assets.anthropic.com/m/ec212e6566a0d47/original/Disrupting-the-first-reported-AI-orchestrated-cyber-espionage-campaign.pdf
2. https://attack.mitre.org/campaigns/C0062/