Part of the AI Regulation News hub.
The FCC is seeking comment on whether artificial intelligence tools should be used to review Universal Service Fund applications, audits and appeals, and what safeguards would have to accompany them
The question is not whether the FCC will use AI internally. It is whether AI should sit in the review of applications, audits and appeals that decide who receives federal universal service support, and the Commission has asked what safeguards that would require without proposing any.
Bottom line: A notice of proposed rulemaking that proposes no AI rule. Within a broader inquiry into how the Universal Service Fund is administered, the Commission asks two clusters of questions about using AI in USF administration, including in the review of applications, audits and appeals.
Who this affects: Universal Service Fund programme participants, being High Cost, Lifeline, E-Rate for schools and libraries, and Rural Health Care; the Universal Service Administrative Company; and anyone advising on automated decision-making in federal benefit and subsidy administration.
Issue date: Published 31 August 2026 at 91 FR 55826. Comments are due on or before 30 September 2026 and reply comments on or before 30 October 2026.
What changed: Nothing yet. What is new is that a federal regulator has put on the record the question of whether AI should review applications, audits and appeals in a subsidy programme, and has named data integrity, governance and quality assurance as the safeguard problem.
Analysis: Note where the AI questions sit. They are not in a technology section. They appear inside the operational efficiency discussion and inside the discussion of deadlines and shot clocks, which is to say they are framed as a remedy for delay. That framing matters, because a tool adopted to reduce turnaround time is measured on turnaround time, and accuracy becomes the constraint rather than the objective.
Primary sources: Maximizing Efficiencies in Universal Service Administration, official GPO text (91 FR 55826) · Same document, GPO PDF
- Instrument
- Maximizing Efficiencies in Universal Service Administration, notice of proposed rulemaking
- Authority
- Federal Communications Commission
- Citation
- 91 FR 55826, pages 55826 to 55835, published 31 August 2026
- Regulatory identifiers
- 47 CFR Part 54; WC Docket No. 26-173; FCC No. 26-52; FR ID 364115
- Jurisdiction
- United States, federal
- Status
- Proposed rule. Open for comment.
- Bindingness
- None. The AI material is a request for comment. No AI requirement, permission or restriction is proposed.
- Issue date / next deadline
- Published 31 August 2026. Comments due 30 September 2026; reply comments due 30 October 2026.
- Scope of the wider inquiry
- Four areas: current USF administration processes, the structure of USF administration and USAC's role, operating costs, and the impact of USAC's Board of Directors
- Predecessor
- USAC Reform Public Notice, DA 26-367, released 15 April 2026
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://www.govinfo.gov/content/pkg/FR-2026-08-31/html/2026-17761.htm
The two places AI appears
The Universal Service Fund supports High Cost, Lifeline, E-Rate for schools and libraries, and Rural Health Care, and is administered day to day by the Universal Service Administrative Company under Commission oversight. This proceeding is about that administration rather than about the programmes themselves.
AI enters the document twice, and in both cases as a question rather than a proposal. The first cluster sits in the discussion of operational efficiency. The Commission asks whether there are operational inefficiencies that could be improved using artificial intelligence, what processes could be improved with AI if any, whether AI should be used to reduce operational turnaround times and costs and in what ways, and whether efficiency in stakeholder engagement would be improved by using AI resources to respond to stakeholder questions.
It then asks the harder questions in the same breath: how privacy and information security concerns should be balanced against the potential benefits, what the financial impact of incorporating AI would be, and, if AI is incorporated, what safeguards need to be put in place to ensure data integrity, governance and quality assurance.
The second cluster is narrower and more consequential. It sits inside the discussion of deadlines and shot clocks for USAC processes, where the Commission asks whether using AI tools to review applications, audits and appeal review processes would help reduce delays while maintaining accurate results.
Why the second question is the one to answer
Application review, audit and appeal are the three points at which a decision is made about a specific participant: whether it receives support, whether funds already disbursed are recovered, and whether an adverse decision stands. A tool placed at those points is not an office productivity measure.
The Commission's own framing supplies the risk. The question appears in a passage about shot clocks, in which the Commission is considering imposing timelines on USAC processes and asking what consequences should follow if a deadline expires. AI is raised there as a way to reduce delay while maintaining accurate results.
That is the classic configuration in which automation is adopted for throughput and evaluated on throughput. If a shot clock creates the pressure and AI is offered as the relief, the measured variable is time to decision, and error rates are a matter for a later audit that the same pressure will also touch.
There is a second-order point specific to this proceeding. The Commission separately notes that under the Payment Integrity Information Act of 2019 and related OMB guidance it is required to implement compliance audits and to report an improper payment rate. Automated review that changes the composition of audit findings would also change that reported rate, which is a statistic with consequences well beyond this docket.
What the Commission has not asked
The document is a request for comment, so absences are not defects. They are nonetheless the gaps a comment can fill, and they are specific.
There is no question about whether a participant would be told that an AI tool was used in reviewing its application, audit or appeal. There is no question about whether a participant could obtain the basis of an automated or automation-assisted determination, or contest it on that ground.
There is no question about the appeal path itself. The Commission asks whether AI could be used in appeal review processes, but an appeal is the remedy for an erroneous first-instance decision, and using the same class of tool at both stages narrows the independence that makes an appeal worth having.
And there is no question about human review. The document does not ask whether a human would make, confirm or be accountable for a determination that an AI tool contributed to. The safeguards question it does ask is framed around data integrity, governance and quality assurance, which are properties of the system rather than of the decision.
How to file, and what will actually land
Comments are due on or before 30 September 2026 and reply comments on or before 30 October 2026, in WC Docket No. 26-173, filed under sections 1.415 and 1.419 of the Commission's rules. The contact given is Stephanie Minnock, Telecommunications Access Policy Division, Wireline Competition Bureau.
The Commission has asked an open question with no proposal attached, which is the stage at which a comment has the most influence and the least to push against. A filing that proposes concrete language, rather than opposing AI in the abstract, is operating in the space the document has actually opened.
The most useful contributions will be specific about which of the three review points is at issue. Using a model to triage a queue of workable applications is a different proposition from using one to score an audit finding or to dispose of an appeal, and a comment that treats them as one thing invites a response that does the same.
One date discipline note. The Federal Register text sets 30 September and 30 October 2026. Where a docket record displays a different date, the published document governs, and this channel has produced that discrepancy before.
The line worth carrying into a filing is the Commission's own: whether using AI tools to review applications, audits and appeal review processes would help reduce delays while maintaining accurate results. That sentence places AI at the three points where a decision is made about a specific USF participant, and it does so inside a discussion about imposing deadlines, which sets throughput as the thing the tool is for. The Commission has already named the safeguards question, asking what would be needed to ensure data integrity, governance and quality assurance, but those are properties of a system. Nothing in the document asks whether a participant is told, whether a determination can be contested on that basis, whether the appeal stays independent of the tool used at first instance, or whether a human remains accountable for the outcome. Comments are due 30 September 2026 in WC Docket No. 26-173, and this is the stage at which proposed language costs least to get adopted.
Source File
https://www.govinfo.gov/content/pkg/FR-2026-08-31/html/2026-17761.htm
Open the GPO text and confirm four things: the header showing 47 CFR Part 54 with WC Docket No. 26-173 and FCC No. 26-52, published 31 August 2026 at page 55826; the DATES line setting comments 30 September 2026 and reply comments 30 October 2026; the operational efficiency passage asking whether inefficiencies could be improved using artificial intelligence and, if AI is incorporated, what safeguards are needed for data integrity, governance and quality assurance; and the Speed of Operations passage asking whether AI tools to review applications, audits and appeal review processes would reduce delays while maintaining accurate results.
Would using artificial intelligence (AI) tools to review applications, audits, and appeal review processes help reduce delays while maintaining accurate results? Federal Communications Commission, Maximizing Efficiencies in Universal Service Administration, 91 FR 55826, 31 August 2026
FAQ
Is the FCC proposing to use AI in the Universal Service Fund?
No. The document proposes no AI rule and adopts no AI tool. It asks two clusters of questions about whether and how AI might be used in USF administration, and invites comment.
What exactly did the FCC ask about applications and appeals?
Whether using artificial intelligence tools to review applications, audits and appeal review processes would help reduce delays while maintaining accurate results. The question appears in the discussion of deadlines and shot clocks for USAC processes.
What safeguards does the Commission mention?
It asks how privacy and information security concerns should be balanced against the potential benefits, and, if AI is incorporated, what safeguards need to be put in place to ensure data integrity, governance and quality assurance. It does not propose any.
Does it ask about telling participants or about human review?
No. The document contains no question about disclosing the use of an AI tool to an affected participant, about contesting an automated or automation-assisted determination on that ground, about keeping appeal review independent of the tool used at first instance, or about human accountability for the outcome.
Which programmes would be affected?
The Universal Service Fund programmes administered by USAC: High Cost, Lifeline, E-Rate for schools and libraries, and Rural Health Care. The proceeding concerns their administration rather than their substantive rules.
When are comments due?
Comments are due on or before 30 September 2026 and reply comments on or before 30 October 2026, in WC Docket No. 26-173.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.