Evidence
State groups from China dominate infrastructure intrusion reporting
| Date | Case | Country | What happened | Source |
|---|---|---|---|---|
| Aug 27, 2025 | Salt Typhoon joint advisory | China | 16 agencies from 13 countries named three Chinese companies | Source |
| Sep 2025 | GTG-1002 AI-run espionage | China | Targets included chemical makers and government agencies | Source |
| 2025 | VOLTZITE moves to Stage 2 | China-linked | SYLVANITE hands access to VOLTZITE for deeper control-system intrusion | Source |
| Jul 8, 2026 | NIS2 infringement cases | EU | Commission took Ireland, Spain, France and the Netherlands to court over late transposition | Source |
Registry
One registry incident involves critical infrastructure compromise
Registry incidents meet a stricter bar than the cases above: public sources name the developer, the failed control and who found it.
| Incident | Date | Agent | Control that failed |
|---|---|---|---|
| Chinese State Group Runs Espionage Through Claude Code | Sep 2025 | Coding agent | Misuse safeguards bypassed by splitting attacks into routine-looking tasks |
Exposure
Four of 17 agent types rate very significant or critical on connectivity surface or physical access
From the attack surface ratings, the agent types where this risk has the widest path in.
| Agent type | Dimension | Rating |
|---|---|---|
| Industrial robots and cobots | Physical access | Very significant |
| Drones | Connectivity surface | Very significant |
| Fleets and swarms | Connectivity surface | Very significant |
| Smart home assistants | Connectivity surface | Very significant |
Controls that bound it
- OT and ICS Integration: how agents connect to industrial control systems
- Monitoring and Anomaly Detection: seeing failures and misuse while they happen
- Access Control and Permissions: what an agent can reach and on whose authority
Rules
Joint guidance covers AI in OT; China raised infrastructure fines tenfold
| Rule | Jurisdiction | What it requires | Status |
|---|---|---|---|
| Principles for the Secure Integration of AI in OT | US and partners | Governance and safety for AI in OT | Dec 3, 2025 |
| America's AI Action Plan | US | AI-ISAC; secure-by-design AI in safety-critical uses | Jul 23, 2025 |
| Amended Cybersecurity Law | China | Infrastructure fines up to RMB 10 million | Effective Jan 1, 2026 |
| NIS2 Directive | EU | Security and incident duties for essential entities | Deadline was Oct 17, 2024 |
Status colors: green in force, blue voluntary standard or framework, amber adopted and phasing in, gray draft or pending, red withdrawn.
Record
Page Record
- Risk group
- Human Risks
- Datasets
- ai_risk_evidence.csv (4 rows); registry incidents (1 linked); ai_attack_surface.csv
- Key sources
- Salt Typhoon joint advisory
GTG-1002 AI-run espionage
VOLTZITE moves to Stage 2
NIS2 infringement cases - Last verified
- October 2026