Strategic PartnersIssue 016 records
OpenAI
First OpenAI edition. OpenAI's GSA OneGov terms, now in their first month, drop the ChatGPT license fee to $0 for state, local and tribal governments, while Plano, Texas standardized on ChatGPT Enterprise through a Texas state contract and budgeted nearly as much for AI governance. An ABA analysis warns that government AI conversations are likely public records. In education, BYU is opening ChatGPT Edu to all students with data limits, OpenAI expanded staff-only ChatGPT for Teachers with a 16-state privacy agreement, and a vendor-co-authored RCT found ChatGPT Edu raised rubric scores while reasoning training added different benefits.
- Evidence records
- 6
- Cross-source patterns
- 2
- Evidence classes
- 2 independent reporting2 vendor claim1 independent research1 academic research
- Outcomes
- 4 emerging1 cautionary1 mixed
- Source freshness
- 3 recent2 undated1 new this fortnight
- Research completed
- 2026-10-30
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
As license costs fall, governance and records work becomes the main rollout cost
OpenAI and GSA removed seat fees for state, local and tribal governments from Oct. 1, while Plano budgeted $150,000 for an AI governance platform against up to $190,000 for ChatGPT licenses, and the ABA analysis shows public-records obligations attach to AI conversations. Interpretation: the binding constraints for SLED ChatGPT rollouts are shifting from license price to governance, consumption control and records handling.
Operating questionHas the agency budgeted for governance, consumption monitoring and records retention, not just licenses?
Supporting evidenceOpenAI; U.S. General Services AdministrationCity of Plano, TexasAmerican Bar Association Forum on Communications Law
Education rollouts pair broad access with data rules and teaching guidance
BYU set data-type limits and a default no-AI coursework rule, OpenAI's K-12 offer is staff-only with a multi-state privacy agreement, and the Bocconi trial found AI access and reasoning training produced different benefits. Interpretation: access alone is not the deliverable; institutions add guardrails and pedagogy.
Operating questionWhat data rules, course policies and instructional supports accompany access?
Supporting evidenceBrigham Young University / Church Educational SystemOpenAIBocconi University; OpenAI Economic Research
Evidence ledger
Full record · every source keeps its link and limitations
Plano, Texas approves up to $390,000 citywide AI package with ChatGPT Enterprise as its standard generative AI platform
Plano City Council approved up to $390,000 for a citywide AI program: up to $190,000 for ChatGPT Enterprise through Carahsoft, starting at 150 licenses estimated at $57,024 a year, plus $150,000 for the Darwin AI governance platform, $10,000 for Microsoft Teams Premium and a $40,000 reserve. The city memo names ChatGPT Enterprise its standard general-purpose platform and pairs it with required training, human review and license reallocation.
Why it matters, evidence and limitations
- Why it matters
- A mid-size city standardizing on ChatGPT Enterprise through a state cooperative contract (Texas DIR) rather than an RFP, and budgeting a separate governance tool at nearly the same cost as the AI licenses, is a concrete local-government procurement pattern.
- Evidence and measured results
- GovTech reports the dollar amounts, license count and consent-agenda approval; the official Plano Legistar file 26-884 memo (also inspected) confirms the amounts, DIR contract numbers, the 150-license minimum and the stated governance controls. No usage or outcome data exist yet.
- Limitations and uncertainty
- Procurement record only; benefits are expected, not measured. The GovTech article date is not shown in the extracted text. The memo does not mention the GSA OneGov $0-license offer effective Oct. 1, 2026, so whether Plano could have bought more cheaply is unresolved.
OpenAI extends GSA OneGov offer to state, local and tribal governments: $0 license fee and 50% off usage from Oct. 1, 2026
OpenAI says that from Oct. 1, 2026 through Dec. 31, 2028, eligible federal, state, local and tribal governments can use ChatGPT with no license fee (normally $15 per user per month), 50% off usage and no minimum, plus Daybreak Blue cyber-defender access at 50% off. GSA independently confirms the 27-month OneGov agreement and state, local and tribal eligibility through direct, reseller and marketplace ordering.
Why it matters, evidence and limitations
- Why it matters
- Now in effect for its first month, this changes the cost basis for every state and local ChatGPT purchase: seat cost drops to zero while consumption becomes the budget variable. It also offers subsidized cyber-defense model access to public-sector defenders.
- Evidence and measured results
- Vendor announcement (OpenAI) corroborated on terms by the GSA press release of the same date. Customer examples (Georgia DOR form digitization, NC Treasurer unclaimed property, CDC literature reviews) are vendor-reported without methods or baselines.
- Limitations and uncertainty
- Vendor claims; usage costs still apply and total cost depends on consumption. Ordering mechanics for state and local buyers using state contracts (as Plano did via Texas DIR) are not explained. Customer outcome figures are not independently verified.
BYU extends CES ChatGPT Edu to the full student body, with data limits and default no-AI rule for coursework
The Church Educational System signed an OpenAI agreement; ChatGPT Edu invitations went to BYU faculty, staff and student employees on Sept. 22, ahead of a planned January 2027 start, and student-body invitations are targeted for late October, possibly November. Data is not used to train models; HIPAA, restricted data and CUI are barred unless the institutional AI Committee approves; students should assume AI is not allowed when an instructor has not said.
Why it matters, evidence and limitations
- Why it matters
- Shows a system-level ChatGPT Edu agreement moving faster than planned and the controls universities attach: data-type limits, committee approval for sensitive data and a default coursework rule.
- Evidence and measured results
- Student newspaper reporting quoting the chair of BYU Operational AI Committee and BYU web guidance; includes faculty and student perspectives. No usage or learning data.
- Limitations and uncertainty
- Private religious university; relies on one administrator and university pages; rollout date may shift; no outcome measures.
OpenAI expands ChatGPT for Teachers to 55 more school systems and introduces a 16-state student data privacy agreement
OpenAI says ChatGPT for Teachers now reaches more than 100 K-12 organizations in 30 states and 300,000+ educators, free for verified U.S. K-12 educators through June 2028 and limited to staff, not students. It introduced a 16-state National Data Privacy Agreement through the Student Data Privacy Consortium and reports 1.9 million time-saving-task messages from Jan. 1 to July 16.
Why it matters, evidence and limitations
- Why it matters
- During the first fall term under the expanded cohort, districts in covered states can evaluate a common privacy agreement rather than negotiating individually, lowering the procurement barrier for district-wide staff AI.
- Evidence and measured results
- Vendor announcement with district leader quotes and OpenAI usage analysis; Skills Jam satisfaction figures (93%, 96%) are self-reported. No independent evaluation of time saved or classroom effects.
- Limitations and uncertainty
- Vendor claims; message counts measure activity, not time saved; first published Aug. 26, 2026 and included here as baseline coverage for this stream first edition during the fall rollout.
ABA analysis: government ChatGPT and other AI chatbot prompts and outputs are likely public records, but retention and cost barriers block access
The article argues AI prompts and outputs used for public business are public records under state law and agency records under FOIA, though no court has ruled. It reports researchers found agencies often claim no responsive records or impose high fees (Colorado quoted $9 million), cites Colorado Springs and HUD withholding, and notes the Army requires retention of prompts while DHS contracted with OpenAI to keep communications as federal records.
Why it matters, evidence and limitations
- Why it matters
- Every state and local ChatGPT Enterprise rollout, including those under the new OneGov terms, creates records-retention and disclosure obligations that procurement and configuration must address.
- Evidence and measured results
- Legal analysis with cited reporting and a preliminary Stanford-led study of 105 records requests; examples include Washington cities and TxDOT producing ChatGPT conversations under state public records laws.
- Limitations and uncertainty
- Commentary by a media-law attorney, advocacy-oriented toward access; not OpenAI-specific; cited study findings are preliminary.
RCT of 1,053 Bocconi undergraduates: ChatGPT Edu raised rubric scores; causal-reasoning training broadened ideas but did not
In a 2x2 randomized trial of 1,053 first-year students, ChatGPT Edu (GPT-4o) access raised expert rubric scores on a business case, while causal-reasoning training increased mechanism identification (about 0.5 SD) and idea diversity without raising rubric scores. The combined condition showed more coherent and falsification logic.
Why it matters, evidence and limitations
- Why it matters
- Gives universities experimental evidence that AI access improves graded output on well-defined tasks but that critical-thinking instruction adds distinct benefits the rubric did not capture, informing how ChatGPT Edu rollouts pair with curriculum.
- Evidence and measured results
- RCT across 13 classes with expert graders, inter-rater reliability reported in the appendix, and OLS with clustered errors; some covariates failed balance checks and were controlled.
- Limitations and uncertainty
- Co-authored with OpenAI employees and contractors; single Italian university and task; GPT-4o rather than current models; publication month only (August 2026); not shown as peer-reviewed.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Independent research
- Research conducted outside the implementing organization.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.