Page last updated October 12, 2026. Superscript numbers on the home and research pages link to the entries below. Each entry names the document, who issued it, and when, and says what GIAG uses it for. We cite the primary document where we have it, and label news reports as news reports.
Statements about named agencies describe what a published review found. They do not assess any agency or vendor beyond that record. See Research Integrity for how GIAG separates research from advisory work and how to send a correction.
Cases cited on the home and research pages
Audit findings, litigation, and vendor disclosures. Where a source is a news report, we say so and will replace it with the primary document when we have it.
-
1
Used forGAO found that CBP did not assess whether an updated set of Automated Targeting System cargo rules was more effective, and had set no performance targets for the rules.
-
2
Opportunities Exist to Improve the Quality of Chat Applications (Report 2026-308-029)
Used forThe IRS began a live chat pilot in November 2017. TIGTA found the IRS had not taken sufficient steps to evaluate the effectiveness of its chat applications, and made nine recommendations that the IRS accepted.
-
3
SBA's Process on Implementing Federal Artificial Intelligence Requirements (Report 26-15)
Used forThe OIG found SBA had not implemented a process to identify high-impact AI uses and had not finalized an AI strategy or governance policies that meet OMB requirements. SBA agreed with all six recommendations.
-
4
Used forReports that TRAX International sued the Army in the U.S. Court of Federal Claims over a White Sands Missile Range award, alleging that AI-generated errors affected the evaluation. The report says GAO denied TRAX’s earlier protest and that the Army has said its source selection authority did not use AI. These are allegations in a pending matter; the report states no ruling. Check the court docket for current status.
-
5
OpenAI Reveals Six Model Incidents Involving Hidden Failures and Unauthorized Uploads
Used forReports OpenAI’s September 17, 2026 disclosure of six incidents of unexpected or concerning model behavior over the prior six months, and cites a Reuters report that related agent activity at Hugging Face began about two months before it was discovered. This entry rests on the news report, not on the disclosure itself.
Background on the Automated Targeting System
What public documents say the system does, and when they said it.
-
6
Privacy Act of 1974; System of Records (DHS/CBP-006, Automated Targeting System), 71 FR 64543
Used forDescribes the Automated Targeting System as screening inbound and outbound cargo, travelers, and conveyances.
-
7
Used forGAO’s description of the Automated Targeting System as a cargo-inspection targeting tool.
-
8
Privacy Impact Assessment for the Automated Targeting System (DHS/CBP/PIA-006(b))
Used forDescribes the system’s modules as of 2012: cargo, passengers, land crossings, and exports.
Law, policy, and procurement
Rules, cases, and enforcement actions that GIAG commentary refers to.
-
9
M-25-21: Accelerating Federal Use of AI through Innovation, Governance, and Public Trust
Used forCurrent OMB direction on federal agency AI governance, including requirements for high-impact AI uses.
-
10
State v. Loomis, 881 N.W.2d 749 (Wis. 2016); Harvard Law Review case comment, 130 Harv. L. Rev. 1530
Used forThe court held that a judge may consider a COMPAS risk score at sentencing, with written cautions, though the vendor treats the tool’s methodology as proprietary. GIAG’s Testing Rigor scoring of public documentation about the tool is published with the working paper. Scores describe documentation, not the tool’s accuracy or fairness.
-
11
FTC Announces Crackdown on Deceptive AI Claims and Schemes (Operation AI Comply)
Used forFederal enforcement against deceptive claims about AI products. Context for the argument that testing claims belong in the contract.
-
12
Used forFederal procurement context for AI contract clauses, including a proposed GSA clause on safeguarding data in large language model systems.
Frameworks GIAG builds on or compares against
Published frameworks. GIAG describes how its pipeline relates to each.
-
13
Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1
Used forThe framework behind Stream One. The framework is voluntary; separate OMB memoranda set requirements for federal agencies.
-
14
Used forGAO’s governance, data, performance, and monitoring framework for auditing AI. A reference point for GIAG’s pipeline.
GIAG working papers and instruments
ThinkCapital’s own publications define the pipeline, the Detection Clock, Testing Rigor, the Human Oversight Quality Index, and the Governance Evidence test. Scoring criteria appear with each working paper.
Read the Working Papers →Help add to the evidence
GIAG is recruiting government practitioners for Phase 2 interviews: 15 conversations by February 1, 2027. The weekly newsletter applies one instrument to a current event.