AI agents used in cyberattack targeting 100 firms, 618K cards exposed
Coordinated AI-Driven Breach Targets Hundreds of Organizations
From September 10–15, 2026, a Chinese-speaking cybercriminal reportedly used AI agents associated with Anthropic's Claude, DeepSeek, and Moonshot AI's Kimi to automate a coordinated attack campaign Forbes. The researcher team, including Snorre Fagerland of NorForge and analysts at Gambit Security, documented the threat actor's targeting of as many as 100 organizations across multiple sectors. The attacker successfully gained access to at least 30 websites during the five-day window, according to reports from cybersecurity researchers Yahoo News Japan.
The scale of exposed financial data is substantial. An exposed server associated with the operation contained information on at least 618,000 payment card records, with approximately 488,000 of those cards linked to U.S. residents Reuters. The available reporting does not establish the precise number of records exfiltrated or confirm fraudulent use of the cards.
Operational Efficiency and Cost Structure
The attack's efficiency underscores the economic advantage AI agents provide to threat actors. Reuters reported that model-token costs for the operation totaled approximately $5,000, translating to roughly $50 per successful compromise. This low operational cost relative to the breadth of access gained illustrates a growing asymmetry in cybercriminal economics: automated agent-driven reconnaissance and exploitation can execute at machine speed with minimal human oversight.
Researchers identified the attacker's use of multiple AI model families—including U.S.-based and Chinese-origin systems—suggesting deliberate diversification to avoid platform-specific detection or rate-limiting. The targeting methodology reportedly leveraged agentic capabilities to automate reconnaissance, credential enumeration, and initial access steps across the victim set.
Open Questions and Verification Gaps
Despite the scale of the reported incident, key details remain unverified. No court filing, criminal charge, arrest, or official law enforcement statement has been publicly confirmed. Anthropic, DeepSeek, and Moonshot AI have not issued official statements regarding the incident or any responsive measures. The precise scope of data exfiltration and the attacker's identity remain undocumented in public records.
The incident reflects a broader trend: as AI agents become more capable and accessible, their dual-use risk—particularly in hands of adversaries with minimal technical overhead—grows. The September breach demonstrates that agentic automation can scale fraud operations beyond traditional manual attack timelines, executing hundreds of intrusions in days rather than months.
The cybersecurity community's documentation of this campaign, led by NorForge and Gambit Security researchers, provides concrete evidence of agent-driven fraud in production, marking one of the largest documented AI-assisted cyberattacks to date.