From the Campus Operations edition of September 29, 2026
Dartmouth faculty evaluation finds substantive AI-characteristic errors in provost Washington Post column
The Dartmouth (Herron and Levin) · Higher education faculty governance / administrative communications integrity · United States (Dartmouth College)
- Publisher
- The Dartmouth
- Original publication
- September 29, 2026
- Source retrieved
- 2026-09-30
- Event date
- 2026-09-29
What happened
On September 29, 2026 Dartmouth professors Michael Herron and Andrew Levin published an independent substantive evaluation of Provost Santiago Schnell Washington Post guest column, arguing that definitive conclusions cannot rest on proprietary detectors alone. They report content failures they consider characteristic of AI-generated text and practically inconceivable from a human measurement expert: a non sequitur shift from academic-integrity framing to capability revelation; misstatement of the Barcaui 2025 study effective sample (abstract n=120 versus 85 retention-test completers underlying key results); omission of convenience-sampling limits and later related studies; near-reproduction of Dartmouth press-release language without citation; and a conflict between acknowledged ChatGPT refinement and The Washington Post attestation that submissions were not created or manipulated with AI. They warn AI-related rule enforcement at Dartmouth may become unworkable without transparent resolution. Authors state AI was used only for background research, not analysis, writing, or editing.
Why it matters
Independent faculty scrutiny of administrative public AI use sharpens campus-operations controls for disclosure, attestation, and quality review of leadership communications. Why now: published September 29, 2026 and not previously archived; complements same-day IHE reporting with content-level evidence.
Evidence and measured results
Attributed faculty guest analysis with specific methodological critiques of one published column. Not a formal audit or peer-reviewed study; personnel process ongoing.
Limitations and uncertainty
Single-essay faculty analysis; campus-ops relevance limited to admin communications governance. Contested personnel matter. Transferability is about policy design, not a general AI accuracy benchmark.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-29; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Administrative AI writing can introduce factual/method errors and false publisher attestations that detectors alone will not catch.
- Stakeholders
- Communications / public affairs
- Provost office
- Faculty governance
- General counsel
- CIO / academic technology
- Discovery questions
- Who verifies cited study methods before leadership op-eds publish?
- What counts as AI manipulation under target publishers rules?
- Are prompt/edit histories retained for contested pieces?
- Is detector output ever used as sole evidence?
- Value hypothesis
- A human methods-and-attestation QC lane for AI-assisted leadership writing can catch abstract/body mismatches and disclosure failures this analysis highlights—without claiming automated perfection.
- Potential engagement
- Ten-day communications QC design producing a pre-publication checklist and attestation workflow for executive external writing.
- Unsupported claims
- Detectors alone prove authorship percentage
- All AI-assisted editing is unethical
- This analysis adjudicates the personnel outcome
Pre-sales engineering
Role takeaway
- Fit
- Quality and attestation controls for AI-assisted administrative public writing.
- Architecture and integration
- Drafting tools with disclosure metadata; citation/source-check workflow; human methods review queue; attestation confirmation before submission.
- Prerequisites
- Publisher rule library
- Access to cited primary sources
- Defined substantial-contribution threshold
- Human reviewer availability
- Constraints
- Single-essay evidence base
- Contested facts in personnel case
- No vendor architecture specified
- Security
- Limit access to draft histories; protect personnel investigation materials; avoid covert monitoring of routine staff AI assistance.
- Proof of value
- Checklist flags an abstract/body sample mismatch in a test draft
- Attestation confirmation recorded before mock submission
- Policy states detectors are insufficient alone
Delivery
Role takeaway
- Work
- Publish checklist, train communications staff, route high-visibility pieces through methods/attestation review, and document outcomes for governance.
- Dependencies
- Communications
- Faculty reviewers
- Provost office
- Counsel for attestation language
- Ownership
- Communications owns the QC lane; provost owns policy thresholds; faculty advisors support methods checks.
- Skills and adoption
- Short training for speechwriters/comms on verifying empirical claims and disclosure language.
- Governance checkpoints
- Policy threshold for mandatory review
- Pre-submission attestation sign-off
- Post-publication correction protocol
- Acceptance criteria
- Checklist adopted for executive external AI-assisted pieces
- At least one dry-run completed
- Detector-only rule prohibited in writing
- Risks
- Selective enforcement
- Review bottleneck delays publishing
- Dispute over what counts as substantial AI contribution
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Admin AI writing stacks need source-citation checks, sample-size/methods verification steps, and attestation gates before external publication—detector scores are not a substitute architecture.
Governance
Who approves, reviews and stays accountable for outcomes?
Shows why role-appropriate rules must define substantial AI contribution, required disclosure, and publisher attestation handling; also why detector-only adjudication is inadequate for high-stakes leadership cases.
Security and privacy
What data, permissions and controls need testing?
Primary risk is integrity and false attestation rather than data breach; still implies need to retain prompt/edit histories for accountable review when leaders dispute AI extent.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Faculty authors model transparent AI-use norms (background research only) while criticizing opaque leadership practice—workforce implication is legitimacy of shared academic standards, not measured staffing change.
Procurement
What should contracts, pricing and exit terms secure?
Discourages procurement narratives that equate AI-detector licenses with authorship governance; favors process/tooling for citation verification and attestation management.
Operating model
Which teams own the service once it runs?
Adds a quality-control operating lane for AI-assisted executive communications alongside broader campus AI adoption programs.
Publication history
- 2026-09-29Campus Operations · Issue 224 resources
Stable resource ID: dartmouth-herron-levin-ai-column-evaluation-20260929