Adversary Intelligence — 2026-09-12

Anthropic Threat Report Documents AI-Enabled State Espionage by Russian, Iranian and Chinese Actors Across Military, Government and Commercial Targets

BLUFAnthropic's report converts theoretical AI-enabled espionage into documented multi-state operational reality, and account bans cannot recall targeting data already in adversary hands.

Anthropic published a 154-page threat intelligence report, its fourth since March 2025, on September 10, covering misuse of Claude models between December 2025 and August 2026 across seven harm areas: cyber operations, influence campaigns, surveillance, scams/fraud, biological misuse, conventional weapons development and illicit distillation 12. The largest cyber case, GTG-20006 ("JackPoterz"), assessed as Russian state-nexus espionage aligned with Midnight Blizzard/APT29, used AI-orchestrated malware, phishing and credential theft against more than 20 government, defense and diplomatic organizations in Ukraine and Europe. It also compromised at least three hotel WiFi vendors to hijack DNS and redirect guest traffic, took over WhatsApp accounts of at least two former senior Ukrainian officials via automation tooling, stole over 300,000 national identity records and commercial registry data on more than 500,000 companies from a North African government technology authority, and used AI agents to autonomously rebuild and redeploy its malware whenever antivirus products detected it 123. A separate Iran-nexus account used Claude to build targeting handbooks against US naval forces in the Middle East, compiling a personnel roster from military photographs, ship/aircraft transponder identifiers, satellite-imagery query scripts, and research into shipboard VSAT and industrial control system vulnerabilities 456. Anthropic also reported a Chinese university-based exploit operation producing over a dozen potential zero-days in a month, a ShinyHunters-linked breach dumping 2,100 cloud tokens across 40 tenants in 34 hours, an intervention against AI-assisted missile-guidance development in northern Yemen, and distillation attacks by Chinese labs including Alibaba, DeepSeek and Moonshot AI, banning the associated accounts and sharing intelligence with government and industry partners 127.

Analysis
Anthropic's disclosure gives the most detailed public documentation to date of state-nexus actors weaponizing commercial AI for espionage, converting theoretical concerns into documented Russian tradecraft at scale via GTG-20006, while a separate Iran-nexus operation built naval-targeting handbooks against US forces through the same platform, indicating overlapping exploitation across at least two state intelligence apparatuses. Account bans close access but do not recover exfiltrated personnel rosters, transponder databases, and vulnerability catalogs already outside Anthropic's control. Anthropic is the sole primary source; independent reporting from CyberScoop and The War Zone draws on the same underlying disclosure. Anthropic's detection capabilities and disclosure incentives alone currently shape the assessed threat surface, and that picture broadens only if competing AI providers follow with comparable disclosures, rather than leaving this as one vendor's competitive and regulatory positioning.

8 sources

Countering misuse of AI: September 2026 (threat intelligence report) - Anthropic
AI lets small actors run state-level hacking campaigns, Anthropic report finds - CyberScoop
Anthropic Reveals How It Stopped 'Kamikaze Drone' Swarms, Biological Weapons And More - Daily Caller
Adversaries Using Claude AI To Target Americans And Develop Missiles - The War Zone
Iran-linked actor used AI to build targeting guides on US Navy - Stars and Stripes
Iran used Claude to target US Navy in Middle East, Anthropic says - Navy Times
Anthropic claims Claude AI used for missile projects, global espionage - Al Jazeera
Anthropic Says Iran-Linked Actor Used Claude to Compile U.S. Navy Targeting Data - gCaptain