Regulatory Compliance

Regulatory compliance

No federal statute specifically governs AI in advancement. Compliance obligations arise from applying existing regulations (FERPA, HIPAA, state privacy laws) to AI use cases, and from emerging state AI governance laws that define advancement-related activities as consequential decisions.

Compliance map by institution type

State law coverage

Hover or tap a state to see its current AI and privacy law obligations for advancement offices. Color indicates enacted law, limited coverage, or pending legislation.

FERPA and AI

No FERPA guidance specifically addresses AI. The Department of Education's Privacy Technical Assistance Center (PTAC) applies existing FERPA principles to AI use cases through vendor FAQ guidance on the school official exception (34 CFR § 99.31(a)(1)(i)(B)).

For an AI vendor to receive education records without student consent under the school official exception, the vendor must: (a) perform a service the institution would otherwise use employees for; (b) operate under the institution's direct control regarding use and maintenance of education records; (c) use records only for the purpose specified in the agreement; and (d) meet the criteria in the institution's annual FERPA notification for school official status.

• AI models do not train on protected student records unless a legitimate educational interest determination is documented
• If a vendor uses education records to train a model that serves other clients, that use exceeds the school official exception scope. Verify that your vendor's AI training practices do not cross this boundary.
• Student opt-outs of directory information are enforced in AI training datasets and output datasets
• AI vendor contracts include FERPA-required provisions for data handling by school officials
• HECVAT 4 assessment is completed for vendors processing student data (includes FERPA alignment questions)

HIPAA and healthcare philanthropy

No published HHS guidance addresses the intersection of AI, HIPAA, and grateful patient fundraising. HHS OCR issued a Dear Colleague letter on January 10, 2025 confirming that Section 1557 nondiscrimination protections apply to AI use by covered entities, effective May 1, 2025.

Grateful patient programs operate in a space where development offices receive limited directory information under HIPAA's treatment, payment, and health care operations (TPO) provisions or through patient authorization. When AI enters this process (e.g., a model scoring patients for philanthropic capacity), the question is whether that processing constitutes a health care operation or exceeds the TPO scope. No OCR guidance addresses this question directly.

• AI-based grateful patient identification does not access protected health information without authorization compliant with 45 CFR § 164.508
• Any AI vendor processing PHI operates under a Business Associate Agreement
• De-identification standards (Safe Harbor per 45 CFR § 164.514(b) or Expert Determination per 45 CFR § 164.514(a)) are applied before clinical data enters AI workflows
• Grateful patient program AI workflows are reviewed by the institutional privacy officer
• AI systems do not use clinical encounter data to infer diagnoses, conditions, or treatments for prospect identification unless the data has been properly authorized or de-identified

New York

New York does not yet have a comprehensive consumer privacy law, but several enacted and pending laws affect AI data processing by advancement offices with NY-based constituents or operations.

SHIELD Act (GBL §§ 899-aa, 899-bb)

Effective since 2020. Applies to any organization holding NY resident PII, including nonprofits. No exemption by size or tax status (a small business safe harbor applies only below 50 employees, $3M revenue, or $5M in assets).

• Inventory all points where PII is collected, stored, processed, or transferred, including AI data flows. When constituent PII enters an AI system, that data flow must be documented.
• AI vendors receiving constituent PII require contractual commitments to reasonable safeguards, not just a terms-of-service checkbox.
• Assess AI-specific risks to PII: model training on PII, data leakage through prompts, retention by third-party AI services, and inference of sensitive attributes from non-sensitive data.
• Maintain data retention and destruction policies that cover PII sent to AI vendors. Indefinite retention by a vendor for model training creates a compliance problem.

NY Health Information Privacy Act (NYHIPA, S-9269)

Passed both chambers June 2026. Awaiting governor's signature. If signed, effective six months after signing. No nonprofit exemption. The prior version was vetoed December 2025.

NYHIPA would regulate "regulated health information" (RHI), defined to include health-related inferences derived through algorithms or machine learning. This provision is directly relevant to advancement: organizations cannot evade regulation by inferring health status rather than collecting it directly. Advancement offices handling health-related constituent data (disability accommodations, wellness program participation, health-related gift designations) would need separate, plainly written authorization for processing. Penalties up to $15,000 per violation with a six-year lookback.

NYC Local Law 144

Effective since 2023. Requires annual bias audits, public disclosure, and candidate notice when automated tools are used for hiring or promotion decisions in NYC. Relevant to institutional HR operations using AI in advancement staffing. Does not apply to donor scoring or prospect research.

State AI and privacy laws

State privacy laws increasingly define AI-based profiling and automated decisions in education, employment, and financial services as consequential or significant decisions, triggering notice, opt-out, and impact assessment requirements. The triggering factor is where constituents reside, not where the institution is located. Institutions with a national constituent base face overlapping obligations across multiple jurisdictions.

Several states do not exempt nonprofits or educational institutions. The states with the highest operational impact for advancement offices:

States requiring immediate attention

• New Jersey (NJDPA, effective January 2025, cure period expired July 2026): Consumer financial information is classified as sensitive data requiring opt-in consent. Prospect research and wealth screening inherently process financial information on NJ residents. Draft regulations require disclosure of how profiling software works and whether it has been evaluated for bias.
• Montana (MTCDPA as amended by SB 297, effective October 2025): Removed the "solely" modifier from profiling opt-out. Consumers can opt out of AI-assisted decisions even when a human reviews the output. This closes the human-in-the-loop defense that most advancement offices rely on.
• Minnesota (MCDPA, effective July 2025; postsecondary institutions: July 2029): Consumers can question automated profiling decisions, demand reasoning behind outcomes, learn what actions would change the result, and request reevaluation after data correction. No other state provides this affirmative challenge right. Covers education and financial services decisions.
• Oregon (OCPA, effective July 2025 for nonprofits): Profiling opt-out covers education and financial services decisions. Global Privacy Control (GPC) signal must be honored as of January 2026. Data protection assessments required for profiling that presents foreseeable risk of harm.
• Maryland (MODPA, effective October 2025): Strict data minimization requirement. Advancement offices must justify each data element collected on MD residents (wealth screening results, real estate records, SEC filings, demographic appends) as reasonably necessary and proportionate to the disclosed purpose. Universities explicitly covered.

Additional state obligations

• Colorado (SB 26-189, effective January 2027): AI scoring that determines which admitted students or donors receive outreach may fall within the "education enrollment or opportunity" category of consequential decisions, triggering notice and disclosure requirements.
• California (CCPA/CPRA ADMT regulations, finalized September 2025): Determine whether Tier 2 or Tier 3 use cases qualify as "significant decisions" in education or financial services. If so, conduct the required risk assessments.
• Connecticut (SB 1103, enacted June 2026): Pre-use notice required when automated processes are a substantial factor in decisions. Must disclose AI-generated content in certain contexts.
• Delaware (DPDPA, effective January 2025): No nonprofit exemption. Profiling opt-out for automated decisions producing legal or significant effects.
• Virginia and Texas: Review opt-out obligations for profiling that produces legal or significant effects.
• Vermont (VDPOSA, effective January 2028): Will require privacy notices to disclose whether personal data is used to train large language models. Profiling challenge rights mirror Minnesota's.

• Map which state laws apply to your constituent base by geography. Most state laws are triggered by constituent residency.
• Document which state law requirements apply to each Tier 2 and Tier 3 use case in the system inventory.
• Review whether your institution honors GPC signals. Oregon and Montana require it for nonprofits.
• Assess whether NJ residents in your prospect research pipeline require opt-in consent for financial data processing.

EU AI Act (for institutions with EU constituents)

AI systems used to determine access to education are classified as high-risk under Annex III, Area 3 of the EU AI Act. This classification carries obligations under Articles 8 through 17, including risk management, data governance, technical documentation, transparency, and human oversight requirements.

Most advancement operations at US institutions will not fall under EU AI Act jurisdiction unless they deploy AI systems that affect decisions about EU residents' access to educational programs, financial services, or employment. Institutions with EU campuses or substantial international enrollment should assess whether their AI use cases fall within scope.

Separate from the high-risk classification, Article 50 transparency obligations apply to any AI system that interacts directly with persons (chatbots must identify themselves as AI) and to AI-generated content (must be marked as AI-generated where technically feasible).

Donor privacy standards

Beyond regulatory requirements, professional association standards and donor expectations impose additional constraints on AI use with constituent data.

• AI-generated solicitation strategies respect donor-stated communication preferences and opt-outs
• Anonymous and confidential gift designations are enforced in AI training data and outputs
• APRA Statement of Ethics principles (accuracy, confidentiality, relevance, donor dignity) apply to AI-generated prospect research outputs
• APRA's Ethics in AI for Fundraising Toolkit recommends ongoing human oversight of AI tools, ethical review of third-party vendors, and attention to bias in training data
• Donors are informed if AI contributes to communications directed to them, per institutional policy and applicable state law