AIR-2026-0064 · DOGE staff used ChatGPT to decide which NEH grants to cancel, per lawsuits over $100M+ in terminations

DOGE staff used ChatGPT to decide which NEH grants to cancel, per lawsuits over $100M+ in terminations

AIR-2026-0064 AIR record · 2026-03-doge-chatgpt-neh-grant-terminations

A real party was affected. This incident caused actual consequences, not a demonstrated capability.

From the supporting source

They uploaded these summaries to ChatGPT, along with this prompt: “Does the following relate at all to D.E.I.? Respond factually in less than 120 characters. Begin with ‘Yes’ or ‘No.’”artforum.com

The record

Disclosed

2026-03-07

Vendor

OpenAI (tool), US DOGE / NEH (deployer)

Product

OpenAI ChatGPT

Risk domain

Agent output quality › Bias and toxicity

Also touches

Unsafe actuation, Confused deputy

System type

Generative model

Who acted

Human acted on output (contested) — ChatGPT classified the grants; DOGE staff executed the terminations

Attack surface

Enterprise assistant

Autonomy level

Assistant

Initial vector

None adversarial

Impact type

Wrongful decision, Financial loss, Discrimination claim

Confidence

High

Submitted by

EAI Internal

Crosswalk

OWASP Agentic Top 10 (2026): ASI09 – Human-Agent Trust Exploitation
MITRE ATLAS: AML.T0048 – External Harms
NIST AI RMF: MEASURE 2.11
EU AI Act: Recital 45
ISO/IEC: TR 24028; 23894
AIUC-1: F. Society
Scope: agent failure
Mitigation: Audit and curate training data for representation; Implement content filtering on inputs and outputs.
Taxonomy: Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery

Blast radius

more than $100 million in withdrawn NEH funding; some projects closed entirely

Capability abused

classification_for_decisionsdelegated_judgement

Defense & remediation

Recorded defense

grant termination review process

Guardrail kinds

Abuse detection

Guardrail outcome

Silent

Fix shipped

litigation ongoing (American Council of Learned Societies, Authors Guild et al. v. DOGE and NEH)

Curator notes

Surfaced via AIAAIC2242. Borderline for scope and deliberately labelled 'decision_delegation_no_tool_use': the model used no tools, but a consequential government decision was delegated to it at scale. Allegations from court filings, not established fact.

Cite as

Enkrypt AI. "DOGE staff used ChatGPT to decide which NEH grants to cancel, per lawsuits over $100M+ in terminations". AIR, AIR-2026-0064. https://agentincident.vercel.app/i/AIR-2026-0064/ (accessed 2026-10-19).