In the wild
Realized harm
Alabama attorney general subpoenas OpenAI over the Hugging Face agent intrusion; 15-state coalition demands OpenAI cease internal cyber evaluations
AIR-2026-0202 AIR record · 2026-08-alabama-ag-openai-probe
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
complete lack of oversight and adequate safeguardsdigit.in
The record
Disclosed
2026-08-24
Occurred
2026-08-24
Vendor
OpenAI
Product
OpenAI frontier model cyber-capability evaluations
Risk domain
Governance › Policy drift
System type
Generative model
Who acted
Agent acted (contested) — records the regulatory fallout of an agent that breached real infrastructure
Attack surface
Autonomous operations
Autonomy level
Fully autonomous
Initial vector
Regulatory response to agent intrusion
Impact type
Regulatory action, Legal exposure
Confidence
Medium
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI01 – Agent Goal Hijack
- MITRE ATLAS
- AML.T0010 – AI Supply Chain Compromise
- NIST AI RMF
- GOVERN 1.5
- EU AI Act
- —
- ISO/IEC
- TR 24028; 23894
- AIUC-1
- E. Accountability
- Scope
- agent failure
- Mitigation
- Implement version pinning for models and prompts; Monitor for behavioral drift.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
OpenAI; 15 US states participating in the demand letter
Capability abused
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
OpenAI says it will publish a technical report and share findings with government authorities
Taxonomy & references
Curator notes
This is the regulatory consequence of 2026-07-openai-huggingface-exploitgym-intrusion, recorded separately because it is the first US state AG enforcement action arising from an autonomous agent incident. AG Steve Marshall issued a subpoena; he and AGs from 14 more states sent Sam Altman a letter demanding OpenAI immediately cease and desist internal cybersecurity evaluations and preserve records. Reuters and SC Media also covered it but neither is quoted here; confidence set to medium because only secondary aggregators are available. OpenAI spokesperson Nate Evans: the Hugging Face incident marked an important moment for AI safety.
Supporting sources
Cite as
Enkrypt AI. "Alabama attorney general subpoenas OpenAI over the Hugging Face agent intrusion; 15-state coalition demands OpenAI cease internal cyber evaluations". AIR, AIR-2026-0202. https://agentincident.vercel.app/i/AIR-2026-0202/ (accessed 2026-10-19).