The scorecard.
Every number on this page is sourced. Every vendor named here has a public DPA, public revenue reporting, or public funder filings behind the line. The argument the campaign makes elsewhere stands on these numbers. Read them first.
§01Headline numbers
The Artificial Intelligence definition presented to the SFPS Board on September 26, 2024 was drafted using ChatGPT. The slide carried this attribution in its footer.19
Slide 3, Board deck of record. The deliberation the board was asked to conduct began with a definition the vendor product helped write. A procedural question about how technical framings reach deciding bodies, not a judgment of any individual.
§02Vendor scorecard
Three vendors are named in the SFPS rollout. Each carries a different economic structure, a different funder profile, a different DPA, and a different long-tail trap. They are not interchangeable.
$53M (FY2018) → $128M (FY2023); ~77% from contributions. Bono-backed, foundation-dependent operating model.
Gates Foundation, Stand Together Trust (Charles Koch network), the Doerr-led Schmidt Futures cohort, the Jeff Yass-aligned Susquehanna Foundation. Schedule B is statutorily redacted; named figures cross-referenced from grant-database disclosures.
Standard Khan Academy DPA permits retention of de-identified learning-pattern telemetry for product improvement. Multi-year retention of interaction logs is typical and not unique to SFPS.
Khanmigo's 'free' positioning rests on a contributor base whose policy preferences are not the district's. The contributor base is the steering wheel; the district is the seat.
~$200M (2023) → $20B+ annualized (2026 reporting). K-12 segment not separately disclosed.
Equity holders and convertible-debt holders include Microsoft, SoftBank, Sequoia, Thrive, Tiger, MGX, and others. The relevant funder is Microsoft via Azure compute exclusivity.
OpenAI Student DPA § 4.2 permits use of de-identified content for 'maintenance, development, support, operation, or improvement' of the service. The 'improvement' word is the long-tail commitment.
ChatGPT for Teachers (the free tier) does not carry a Student DPA by default. A teacher who pastes a student's writing into the free product may be triggering a FERPA violation. FERPA enforcement against a district can run $15K–$75K per incident plus loss of federal funding.
Acquired Istation in June 2024 in a Macquarie-facilitated transaction. Combined footprint ~1,800 districts; contracts in 13 other states. Voice-recording corpus is the principal data asset.
Owens Family Foundation; Owl Ventures; Reach Capital; post-merger ownership reshaped by Macquarie Capital. Investor base is education-specialist VC; less ideological than Khan's, more financial.
Amira's standard DPA is the strongest of the three on substance: no external LLM inference on student voice, no model training on identified student data. That baseline now carries three July 2026 qualifications, each on the record: Portland's $5,160 pilot contract had checked boxes for student disability status, migrant status, and living situation — Amira says no such data was transmitted; why the boxes were checked went unanswered. Asked by the House Education Committee chair where New Mexico student data is stored, the CEO offered assurances without specifics. And the privacy policy itself changed on July 20 — some terms tightened, others narrowed; the version-by-version record is at /advocacy/sfps/amira-policy-diff. The SOC 2 and 1EdTech certifications the company cites are vendor-asserted: no auditor is named and no public directory listing was located.
Voice corpus retention is the long-tail asset. A K-2 child read aloud for two years builds a multi-thousand-utterance fingerprint. The DPA governs use; it does not govern eventual sale of the company. PED's own guidance places one control locally: district administrators can elect to remove retention of student voice recordings.
| Vendor | Named reviewer | Reviewer's funding ask |
|---|---|---|
| Khan Academy | Common Sense Media; TeachAI policy briefs | Common Sense Media solicited $100M from AI vendors (POLITICO, Apr 1, 2026).20 TeachAI is the vendor coalition's policy arm. |
| OpenAI | Common Sense Media; TeachAI policy briefs | Same as above. OpenAI is one of the vendors named in the POLITICO reporting.20 |
| Amira Learning | Common Sense Media; education-press reviews | Same reviewer; Amira not named in the POLITICO solicitation reporting. |
The reviewer the field reads as independent is asking the vendors under review for nine figures. A district that takes a "Common Sense" or TeachAI rating as third-party validation is reading a rating funded, in part, by the rated company.
The three vendors above are the named ones. The Legislature’s analyst reports that New Mexico has no formal evaluation process for AI education tools25 — which means no one, including this page, can currently enumerate the AI and adaptive-learning tools actually running on district devices. That is not a claim about any unnamed vendor. It is a statement about the absence of the list. The public-records request on the IPRA page asks for the list.
A New Mexico jury orders Meta to pay $375 million for violating state consumer-protection law — the first state in the nation to prevail at trial against a major technology company for harming young people.26
The Legislature’s education analyst reports that New Mexico “does not currently have statewide procurement standards for AI tools, student data privacy requirements specific to AI, and a formal evaluation process for AI education tools” — and that it is unclear what vetting PED used before mandating a voice-recording AI assessment for every K-2 student in the state.25
The same state holds both facts. The asks on the proposal page are what it looks like for a district to reconcile them.
§03The NC DPI laundromat
The governance scale SFPS adopted for AI in classrooms was developed by the North Carolina Department of Public Instruction. Importing a framework is not, by itself, a problem. The problem is that the import functions as a deliberation laundromat. A scale developed in one state by one set of authors arrives in New Mexico already carrying the legitimacy of a peer state's seal, which lets the district adopt a policy product without re-running the deliberation that would have made it the community's own.15
The question for the board is whether the imported framework had its New Mexico discount applied — for the broadband floor, for tribal consultation, for Spanish-language families, for the cohort-decade horizon — before it became district policy.
A third-party framework is not a district policy. The path the field has run for two decades is: industry funds the studies, industry-funded researchers populate the task forces, task forces author the principles, districts adopt the principles as guidance, and the guidance becomes a procurement justification. TeachAI's Seven Principles, which the Sep 26 2024 deck names as the SFPS framing, is the rail SFPS rode in on. The Board has not voted on those principles. The Board has not authored district policy that constrains the principles. Until it does, what the district has is a vendor-coalition framework, not a community policy.
§04The broadband floor
An AI tool is a service running on top of a stack of household infrastructure. Where the stack is thin, the tool is not what the brochure says it is. The numbers below are the floor in New Mexico right now.
§05The undercount
High-end estimated share of Native K-12 students undercounted in New Mexico district-level reporting due to categorical, residential, or self-identification mismatches. An AI rollout that is evaluated on data which undercounts its largest at-risk population by this much cannot, by construction, surface its own failure mode for that population.
§06Sources
- Khan Academy, Khanmigo district pilot reporting, 2024–2025. Engagement defined as the share of assigned students using the tool at least once per week.
- Independent K-12 AI tutor evaluation aggregated across districts, 2025. End-of-unit assessment as the common outcome measure; no statistically significant difference observed between AI-tutor and general-web-search conditions.
- Microsoft Research, Tutor CoPilot field trials, 2024–2025. Reported effect ~4 percentage points on the studied learning outcome in the better-served arm. Effect attenuates in arms with thinner home setup.
- Aggregated teacher self-report from US K-12 surveys, 2024 and 2026 waves. Definition of 'use' varies across surveys; the trend direction is consistent across sources.
- US Census American Community Survey and FCC Form 477 reporting, 2024–2025. Threshold defined per the current FCC broadband definition. The figure is the share of K-12 enrollment without reliable in-home service at threshold.
- OpenAI 2026 revenue reporting via The Information and Reuters. Annualized run-rate, not GAAP revenue. K-12 segment not separately disclosed in any public filing.
- Khan Academy IRS Form 990, FY2018 and FY2023. Total revenue and contributions percentage are line items 1 and 1f respectively.
- Cross-reference of Gates Foundation, Stand Together Trust, Schmidt Futures, and Susquehanna Foundation grant databases (2018–2025) for confirmed Khan Academy awards. Schedule B contributor identity is not publicly disclosed by Khan; named figures are reconstructed from donor-side filings.
- Khan Academy Khanmigo Data Processing Agreement, current public version. Retention and de-identification language governs the points cited.
- OpenAI revenue trajectory aggregated from The Information, Reuters, and Bloomberg reporting, 2023–2026.
- OpenAI Student Data Processing Agreement, current public version, § 4.2. The 'maintenance, development, support, operation, or improvement' clause is the operative language.
- 20 U.S.C. § 1232g (Family Educational Rights and Privacy Act). Enforcement remedies include loss of federal funding and Department of Education administrative actions; per-incident civil penalty estimates vary by published guidance.
- Amira Learning / Istation merger announcement, June 2024, Macquarie Capital as facilitator. District footprint figures from combined company disclosures.
- Amira Learning current Data Processing Agreement, public version. The provisions on external LLM inference and model training on identified student data are the points cited.
- North Carolina Department of Public Instruction, Generative AI Implementation Recommendations and Considerations for PK-13 Public Schools, January 2024.
- New Mexico Office of Broadband Access and Expansion, county-level household broadband adoption release, 2024–2025. Lowest-served-county broadband adoption (share of households with a broadband subscription of any kind): Hidalgo 2.1%, Socorro 5.7%, Colfax 10.2%, Catron 14.6%.
- FCC and Navajo Nation Telecommunications Regulatory Commission, 2024–2025. Estimated share of Navajo Nation households below 10 Mbps service is 40–100% depending on chapter geography and surveying method. FCC's tribal-area broadband gap figure is 68%.
- Cross-source aggregation of NM Public Education Department reporting, US Census American Community Survey 5-year tribal estimates, and tribal-government enrollment data, 2024–2026. The ~70% high-end figure reflects estimated undercount of Native students in district-level K-12 reporting due to categorical, residential, or self-identification mismatches.
- Santa Fe Public Schools, Board of Education Study Session: Artificial Intelligence in SFPS, presented by Chavez, Romero, Weaver, and Lewis, September 26, 2024. Slide 3 footer: 'Definition of Artificial Intelligence provided by ChatGPT, adapted by Dr. Neal Weaver, Chief Information and Strategy Officer, Santa Fe Public Schools.'
- POLITICO, April 1, 2026, reporting on Common Sense Media's $100M solicitation from AI vendors. Two of the three vendors named in the SFPS rollout (Khan Academy and OpenAI) sit inside that funding network.
- New Mexico Public Education Department, AMIRA-ISIP Guidance 8.2025: Amira ISIP required K-2 statewide (grade 3 optional), continued under the state's Istation contract after the merger ('Istation has been rebranded as Amira'). The local retention control, verbatim: 'If a district or state charter does not wish for students' audio to be retained, local administrators can elect to remove the retention of the student's voice.'
- Willamette Week, 'AI Company Tells Portland School Board It Does Not Collect Student Data,' July 4, 2026. Contract obtained by public records request; COO response letter dated June 29, 2026; the paper reports the company did not answer follow-up questions about why the data boxes were checked.
- Legislative Education Study Committee meeting, Taos, July 15, 2026, Item 6: Update on Assessments and Amira (agenda, nmlegis.gov). Hearing coverage carries the CEO's data-storage answers and PED's confirmation that the assessment contract opens for competitive bids in school year 2027-28.
- Amira Learning privacy policy, April 12, 2026 (Wayback Machine capture) and July 20, 2026 ('Last Updated' date) versions, compared section by section at /advocacy/sfps/amira-policy-diff. The July 20 version is archived on this site.
- Ortiz, S., 'AI in K-12 Education: State Policy Trends and Implications for New Mexico,' LESC Policy Brief, May 27, 2026. The quoted sentence appears on page 4; the brief also reports it is 'unclear what vetting processes and stakeholder input PED used to implement Amira ISIP.'
- New Mexico Department of Justice press release, 'New Mexico Department of Justice Wins Landmark Verdict Against Meta,' and CNBC trial coverage, March 24, 2026: $375 million in civil penalties for violations of New Mexico consumer-protection law; the first state to prevail at trial against a major technology company for harming young people. Injunctive-relief proceedings followed.
- Report: LESC Artificial Intelligence Working Group (Bedeaux), July 23, 2025, nmlegis.gov. Recommends minimum data privacy and security standards, a vetting process for AI tools, prohibitions on training from sensitive student data, tribal data sovereignty protections, and a clear opt-out procedure.
The source bank at /advocacy/sfps/sources carries verbatim excerpts as they are confirmed. If a number here is wrong, write the campaign at kinetic-trust@proton.me with the source and the correction will land within forty-eight hours.