The Moonlighting Gap in Campus AI Governance

Four Policies, One Gap None of Them Reach

Maryland’s public colleges let individual faculty set their own AI policy for their own classrooms, with no system-wide standard behind it GovTech. The University of Michigan went the other way, issuing six institutional guidelines that apply across campus regardless of instructor preference GovTech. State legislatures are writing AI restrictions directly into K-12 law Newsweek, and at City Colleges of Chicago, AI terms were serious enough to be part of what averted a strike during contract bargaining CBS News.

Four authorities, four altitudes, same institutions, none deferring to the others. That jurisdictional overlap is real, but it is not the part of this story that compliance leaders are missing. The costlier gap sits underneath all four layers at once.

The Vendor Dependency Sitting on Top

Higher ed’s parallel debate over tech sovereignty, whether institutions can reduce dependence on a handful of large AI vendors for the infrastructure underneath campus policy, is not separate from this Inside Higher Ed. The same vendors institutions are trying to gain distance from are often the ones buying labor downstream, including from the people those institutions employ.

The Workforce No Policy Layer Covers

Platforms like AITraining.jobs list remote roles for credentialed teachers working as AI trainers and evaluators across subjects including math and STEM, with pay ranging from $50 to $135 an hour, a reported median of $50, and rates up to $400 an hour at Surge AI AITraining.jobs. That is professional judgment developed on a public payroll, sold in the evening to improve a commercial model in the same field. Commercial skills-academy products, like WorkforceEdge’s packaged cloud and AI certification offering, operate in the same uncovered space, selling direct to employers outside any of the four governance layers above WorkforceEdge. None of a faculty member’s classroom policy, a university’s institutional guideline, a state’s K-12 statute, or a union’s bargained contract language was written with this kind of moonlighting in mind, because none of those instruments was built to look past the classroom door.

What the Frameworks Actually Ask For

This is the exact seam NIST’s AI Risk Management Framework and ISO 42001 are built to close, and the gap is specific, not generic. NIST’s RMF defines “AI actors” across the full lifecycle, not just the vendor selling a finished product, and its GOVERN function requires organizations to assign accountability for risk arising from third-party contributions to a model, including data and evaluation labor. ISO 42001’s supplier and third-party relationship controls require an AI management system to document who is actually doing the labeling, grading, and evaluation work upstream of deployment. Applied here, that means an institution’s AI governance posture is incomplete if it cannot account for its own credentialed staff supplying that labor to outside vendors. In practice, the only instrument already positioned to catch this is not a new AI policy at all. It is the outside-employment and conflict-of-interest language most institutions already have, almost none of which was written with a teacher’s AI-training side income in mind.

Compliance leaders drafting policy at any of the four altitudes should ask a fifth question none of the current layers asks: does this policy, contract, or statute say anything about what a credentialed employee is permitted to do for a commercial AI company on their own time, using the expertise that employment paid to build.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The core argument—that faculty moonlighting as AI trainers creates a governance gap unaddressed by existing policy layers—is coherent and novel, but the claim that NIST RMF and ISO 42001 require insti
Source & Claim Verification	glm-5.3-flash-fp4 · local	held. The news anchors and pay figures check out against their sources (GovTech, CBS, AITraining.jobs all match), but the entire NIST AI RMF / ISO 42001 paragraph makes specific claims about what those fram
Regulatory & Framework Fidelity	Mistral · fallback (mistral-medium-3.1)	cleared. The briefing accurately identifies the NIST AI RMF and ISO 42001 gaps in addressing third-party AI labor contributions but does not explicitly tie these to enforceable controls or audit mechanisms.
Technical Accuracy	Llama	cleared. The article accurately highlights the gap in AI governance related to faculty moonlighting as AI trainers for commercial vendors, and correctly references relevant frameworks like NIST’s AI Risk Manag
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies a novel gap in AI governance and proactively addresses potential counterarguments by framing the issue within existing compliance frameworks and institutional polic
Novelty & Non-Duplication	Grok	cleared. The faculty/teacher moonlighting-as-AI-trainer gap and the COI/outside-employment lever are not in the back catalogue, despite familiar four-layer campus-policy and vendor-sovereignty framing.
Validation	DeepSeek	cleared. The central claim that existing campus AI policies do not address faculty moonlighting as AI trainers is validated by the provided sources, which show the policies’ scope is limited to institutional u

Sources cited: 17. Validation challenges: 0. Learn how these briefings are written and verified.