Pennsylvania — AI Duty Profile (Act 35 of 2025 Digital Forgery; Act 125 of 2024; Insurance AI Notice 2024-04; Insurance Data Security Act; BPINA; DOS AI Enforcement Task Force): AI Compliance Requirements

Do not generate or distribute, as genuine, an AI-fabricated voice, image or video of a real, identifiable individual with intent to defraud or injure anyone — and do not knowingly supply such material to someone else who will. This covers AI voice cloning for impersonation fraud, deepfake imagery for financial scams, and fabricated audio/video "evidence". The intent to defraud or injure is an ELEMENT of the offense: consented or clearly-labelled synthetic media is not digital forgery. Grading rises from a first-degree misdemeanor to a third-degree felony where the forgery is committed through involvement in a scheme to defraud, coerce or commit theft of monetary assets or property. Effective September 5, 2025.

Act 125 of 2024 (SB 1213, signed 2024-10-29) extended two existing Crimes Code offenses to synthetic media. 18 Pa.C.S. § 3131 (unlawful dissemination of intimate image) now reaches "an artificially generated sexual depiction of an individual" disseminated with intent to harass, annoy or alarm; 18 Pa.C.S. § 6312 (sexual abuse of children) now reaches AI-generated child sexual abuse material, including intentionally viewing, possessing or controlling it. Any organization operating image- or video-generation capability accessible to users must have technical controls (prompt/output filtering, known-CSAM detection, age controls) and an incident path — the exposure is criminal, not regulatory, and does not depend on a commercial motive.

Notice 2024-04 Section 1 states the rate-act position directly: the Casualty and Surety Rate Regulatory Act, the Fire, Marine and Inland Marine Rate Regulatory Act, the Property and Casualty Filing Reform Act, Article VII of the Workers' Compensation Act and the Title Insurance Companies article "require that property/casualty (P/C) insurance rates not be excessive, inadequate or unfairly discriminatory. The requirements of these acts apply regardless of the methodology that the insurer used to develop rates, rating rules and rating plans subject to those provisions." The notice then places the burden explicitly: "an insurer is responsible for assuring that rates, rating rules and rating plans that are developed using AI techniques and predictive models that rely on data and machine learning do not result in excessive, inadequate or unfairly discriminatory insurance rates with respect to all forms of casualty insurance—including fidelity, surety and guaranty bond—and to all forms of property insurance—including fire, marine and inland marine insurance, and any combination of any of the foregoing." This is the one place in the instrument where a BINDING statutory standard, not a guideline, is applied to AI output: the rate acts bind on their own force and the notice merely confirms that model-derived rates are inside them.

The substantive floor under the whole Pennsylvania insurance-AI surface. Notice 2024-04 Section 1 states that the Unfair Insurance Practices Act "regulates trade practices in the business of insurance by defining practices that constitute unfair methods of competition or unfair or deceptive acts and practices and prohibiting the trade practices so defined or determined", and that the Unfair Claims Settlement Practices regulations "set forth standards for the investigation and disposition of claims arising under policies or certificates of insurance issued to residents in this Commonwealth." The operative sentence is methodology-blind: "Actions taken by insurers in this Commonwealth must not violate the UIPA or the UCSP regulations, regardless of the methods the insurer used to determine or support its actions." The notice then states what insurers are expected to do about it: "insurers are expected to adopt practices, including governance frameworks and risk management protocols, that are designed to ensure that the use of AI systems does not result in: 1) unfair trade practices, as defined in the UIPA; or 2) unfair claims settlement practices, as defined in the UCSP regulations." Section 3 restates the standard for AI decisions specifically — they must not be "inaccurate, arbitrary, capricious or unfairly discriminatory" — and the notice defines "[a]dverse consumer outcome" as "[a] decision by an insurer that is subject to insurance regulatory standards enforced by the Department that adversely impacts the consumer in a manner that violates those standards." Practical effect: an AI-driven claim denial, delay, or settlement practice is judged against the same UCSP standards as a human one, with no allowance for model opacity.

R524 STANDING UTILIZATION-REVIEW SWEEP — Pennsylvania makes it five for five (Illinois, Massachusetts, Virginia, Maryland, now Pennsylvania: every state whose insurance-AI surface has been built to premium depth also carries a licensed-clinician adverse-determination rule that constrains AI far harder than the AI guidance does). Article XXI of the Insurance Company Law of 1921, added by Act 68 of 1998 (the Quality Health Care Accountability and Protection act, 40 P.S. §§ 991.2101—991.2193), provides at 40 P.S. § 991.2152 that "[u]tilization review that results in a denial of payment for a health care service shall be made by a licensed physician" meeting the qualifications in section 2155(d). Narrow professional exceptions exist and are themselves limited: "[a] licensed psychologist may perform a utilization review for behavioral health care services within the psychologist's scope of practice if the psychologist's clinical experience provides sufficient experience to review that specific behavioral health care service", but a psychologist may not "review the denial of payment for a health care service involving inpatient care or a prescription drug"; a licensed dentist may likewise review dental services within scope. The section also prohibits incentive structures: "[c]ompensation to any person or entity performing utilization review may not contain incentives, direct or indirect, for the person or entity to approve or deny payment for the delivery of any health care service." THE AI CONSEQUENCE: an AI or algorithmic system is not a licensed physician, psychologist or dentist. It may triage, surface evidence, flag cases and draft rationales, but the adverse determination itself must be MADE by the qualified licensed clinician, who must actually exercise judgment rather than ratify a model output at volume. A throughput pattern in which a clinician approves model-generated denials faster than review is possible is the exact fact pattern regulators and plaintiffs have attacked in other states.

Pennsylvania is enforcing existing professional-licensing law against AI products directly. On 2026-02-27 the Governor announced a 12-member AI Enforcement Task Force and a public complaint process at the Department of State, whose stated remit is to evaluate whether AI bots are engaged in UNLICENSED PROFESSIONAL PRACTICE under existing law, with the Attorney General handling the parallel Unfair Trade Practices and Consumer Protection Law track. The Commonwealth then sued Character Technologies, Inc. over chatbots that presented as licensed medical professionals — one bot claimed to be a doctor of psychiatry and supplied an invalid Pennsylvania medical licence number. Concrete controls: bot names, avatars, personas and profiles must not assert professional credentials; outputs must not diagnose, prescribe or otherwise perform a licensed activity; the system must never emit a licence number or claimed licensure; and disclaimers must be strong enough to survive the design choices around them, since a boilerplate disclaimer beneath a persona built to look like a clinician is exactly what this action attacks.

Document your organization's policy on use of AI voice synthesis, image generation, and video synthesis technologies. Establish clear prohibitions on generating non-consensual synthetic media of real individuals for harmful purposes.

Pennsylvania adopted the NAIC model AI bulletin as Notice 2024-04 on 2024-04-06 (the eighth state to do so). Every insurer holding a certificate of authority or otherwise authorised to do insurance business in Pennsylvania — including HMOs, PPOs and health plan corporations — is expected to maintain a written program for the responsible use of AI systems ("AIS program"), covering: risk-management controls and internal audit; senior-management accountability to the board or a board committee; documented compliance; controls across the full AI lifecycle from design to retirement; data governance and bias analysis; and due diligence, contractual audit rights and ongoing oversight of third-party AI vendors and their data. The Department may request the governance framework, model-specific documentation, data lineage, validation testing results and third-party contracts in any investigation or market-conduct examination.

Notice 2024-04 Section 3 § 1.3 expects the AIS program to "vest responsibility for the development, implementation, monitoring and oversight of the AIS program and for setting the insurer's strategy for AI systems with senior management accountable to the board or an appropriate committee of the board." Group 2.0 (Governance) then expects the framework to "prioritize transparency, fairness and accountability in the design and implementation of the AI systems, recognizing that proprietary and trade secret information must be protected", and to address: § 2.1 the policies, processes and procedures — including risk management and internal controls — to be followed at each stage of an AI system life cycle "from proposed development to retirement"; § 2.2 the requirements adopted to document compliance, expressly "developed with Section 4 in mind" (i.e. authored so the examination production list can be answered); and § 2.3 the internal accountability structure, itemised as (a) centralized, federated or otherwise constituted committees drawn from business units, product specialists, actuarial, data science and analytics, underwriting, claims, compliance and legal; (b) scope of responsibility and authority, chains of command and decisional hierarchies; (c) "[t]he independence of decision-makers and lines of defense at successive stages of the AI system life cycle"; (d) monitoring, auditing, escalation, and reporting protocols and requirements; and (e) ongoing training and supervision of personnel. Programme reach is set by § 1.6 (all AI systems across "product development and design, marketing, use, underwriting, rating and pricing, case management, claim administration and payment, and fraud detection") and § 1.7 (all life-cycle phases including "design, development, validation, implementation (both systems and business), use, on-going monitoring, updating and retirement"). § 1.5 permits the programme to sit inside or outside the existing Enterprise Risk Management programme and to adopt a third-party framework, naming "the National Institute of Standards and Technology Artificial Intelligence Risk Management Framework, Version 1.0". § 1.8 extends the programme to AI "whether developed by the insurer or embedded within an affiliate or third-party vendor process".

Notice 2024-04 Section 3 group 3.0 expects the AIS program to "document the insurer's risk identification, mitigation, and management framework and internal controls for AI systems generally and at each stage of the AI system life cycle", addressing seven items. § 3.1: the oversight and approval process for developing, adopting or acquiring AI systems, "as well as the identification of constraints and controls on automation and design to align and balance function with risk". § 3.2: data practices and accountability procedures, "including data currency, lineage, quality, integrity, bias analysis and minimization, and suitability". § 3.3: management and oversight of predictive models and the algorithms used in them, comprising (a) inventories and descriptions of the predictive models, (b) "[d]etailed documentation of the development and use of the predictive models", and (c) assessments such as "interpretability, repeatability, robustness, regular tuning, reproducibility, traceability, model drift and the auditability of these measurements where appropriate". § 3.4: "[v]alidating, testing and retesting as necessary to assess the generalization of AI system outputs upon implementation, including the suitability of the data used to develop, train, validate and audit the model" — the notice specifies the accepted method as "comparing model performance on unseen data available at the time of model development to the performance observed on data post-implementation, measuring performance against expert review or other methods." § 3.5: protection of non-public information, particularly consumer information, "including unauthorized access to the predictive models themselves". § 3.6: data and record retention. § 3.7: for predictive models specifically, "a narrative description of the model's intended goals and objectives and how the model is developed and validated to ensure that the AI systems that rely on such models correctly and efficiently predict or implement those goals and objectives." Governance § 2.4 pairs with this, expecting documented processes "for designing, developing, verifying, deploying, using, updating and monitoring predictive models, including a description of methods used to detect and address errors, performance issues, outliers or unfair discrimination in the insurance practices resulting from the use of the predictive model."

Notice 2024-04 Section 3 group 4.0 expects each AIS program to address the insurer's process for acquiring, using or relying on "(i) third-party data to develop AI systems; and (ii) AI systems developed by a third party", through standards, policies, procedures and protocols covering three items. § 4.1: "[d]ue diligence and the methods, which may include human oversight, employed by the insurer to assess the third party and its data or AI systems acquired from the third party to ensure that decisions made or supported from such AI systems that could lead to adverse consumer outcomes will meet the legal standards imposed on the insurer itself" — the operative principle being that buying the model does not move the legal standard off the insurer. § 4.2: where appropriate and available, contract terms that (a) "[p]rovide audit rights or entitle the insurer to receive audit reports, or both, such as System and Organization Control 2 reports or other generally accepted reports, performed by qualified auditing entities, which specifically encompass the AI in scope of the review", and (b) "[r]equire the third party to cooperate with the insurer with regard to regulatory inquiries and investigations related to the insurer's use of the third party's product or services." § 4.3: actually exercising those rights — "[t]he performance of contractual rights regarding audits or other activities, or both, to confirm the third party's compliance with contractual and, where applicable, regulatory requirements." Two contracting details carry disproportionate weight: a generic SOC 2 report does not satisfy § 4.2(a) unless it "specifically encompass[es] the AI in scope of the review", and § 4.3 makes an unexercised audit right a gap rather than a control.

Notice 2024-04 Section 1 makes the CGAD filing an AI-reporting channel: Chapter 39 of Title 40 "requires insurers to report on governance practices and to provide a summary of the insurer's corporate governance structure, policies and practices", and — the operative sentence — "[t]he CGAD requirements are applicable to the elements of the insurer's corporate governance framework that address the insurer's use of AI systems to support actions and decisions that impact consumers." This matters because it is a PERIODIC, SELF-EXECUTING disclosure obligation rather than a contingent one: unlike the Section 4 examination production list, which only bites when the Department opens an investigation or market conduct action, the CGAD comes due on its own cycle and an insurer whose AI governance is undocumented has to file something about it anyway. Insurers that stood up an AIS program under § 1.3 with senior management accountable to the board have the CGAD content already; those that did not will find the gap surfaced in a filing signed at officer level.

Notice 2024-04 Section 4 is a production list, and it is the part of the instrument with operational teeth. It warns that "[r]egardless of the existence or scope of a written AIS program, in the context of an investigation or market conduct action or at any time determined necessary by the Insurance Commissioner, an insurer can expect to be asked to respond to an inquiry or provide documentation, or both" — so having no AIS program does not reduce what must be produced. Group 1.0 (AI system governance, risk management and use protocols) itemises: § 1.1(a) the current written AIS program; (b) documentation evidencing its adoption; (c) "[t]he scope of the insurer's AIS program, including any and all AI systems and technologies whether or not included in or addressed by the AIS program" — note that AI deliberately left OUTSIDE the programme is still discoverable; (d) the structure and mechanisms by which the programme is tailored and proportionate; (e) policies, procedures, guidance and training materials, expressly including (i) development/adoption/acquisition processes covering constraints and controls on automation and data governance and controls "including practices related to data lineage, quality, integrity, bias analysis and minimization, suitability, and data currency", (ii) predictive-model management processes "including measurements, standards, or thresholds adopted or used by the insurer", and (iii) protection of nonpublic information "including unauthorized access to predictive models themselves". § 1.2: pre-acquisition/pre-use diligence, monitoring, oversight and auditing of third-party data or AI systems. § 1.3: evidence of implementation and COMPLIANCE with the programme, including (a) formation and ongoing operation of coordinating bodies, (b) data practices and accountability documentation, (c) model and AI system inventories plus, for any model under investigation, documentation of compliance with programme policies, data source, provenance, lineage, quality, integrity, bias analysis and minimization, suitability and currency, and the techniques, measurements and thresholds used, and (d) validation, testing and auditing documentation "including evaluation of model drift". Group 2.0 adds, where third parties are involved: § 2.1 due diligence conducted on third parties and their data, models or AI systems; § 2.2 the vendor contracts themselves, "including terms relating to representations, warranties, data security and privacy, data sourcing, intellectual property rights, confidentiality and disclosures, and/or cooperation with regulators"; § 2.3 audits or confirmation processes performed regarding third-party compliance; and § 2.4 validation, testing and auditing documentation including model-drift evaluation.

The Insurance Data Security Act (40 Pa.C.S. Ch. 45, added by Act 2 of 2023, effective 2023-12-11) applies to every licensee — insurers, third-party administrators, producers and rating organizations. Licensees must run a risk assessment, maintain a written information security program with corporate oversight, certify compliance annually (domiciled insurers), investigate cybersecurity events, and notify the Insurance Commissioner as promptly as possible and no later than FIVE BUSINESS DAYS. The provision that matters most for AI adoption is the third-party service-provider oversight duty, which became effective 2025-12-11: an AI vendor processing nonpublic information for a licensee is a third-party service provider and must be diligenced and contractually bound accordingly. Phased dates from the Department's own guidance: risk assessment / security program / corporate oversight by 2024-12-11; third-party service-provider oversight by 2025-12-11.

The Breach of Personal Information Notification Act reaches any State agency, political subdivision, or individual or business doing business in the Commonwealth that maintains, stores or manages computerised personal information of Pennsylvania residents — no size threshold. Act 33 of 2024 (P.L. 427, No. 33, signed 2024-06-28, effective 2024-09-26) tightened it: notice to affected individuals without unreasonable delay (government entities within 7 business days); where more than 500 individuals must be notified, simultaneous notice to the Office of Attorney General — giving organization name and location, breach date, incident summary, estimated total affected and estimated Pennsylvania residents affected — and to the nationwide consumer reporting agencies (the CRA threshold dropped from 1,000 to 500); and where the breach involves Social Security, driver's licence/state ID or financial account numbers, 12 months of no-cost credit monitoring plus a complimentary credit report. Encryption is a safe harbour unless the key was also compromised. This is the operative Pennsylvania duty for AI vendors and AI-using businesses holding Pennsylvania customer or employee data — training corpora and prompt logs included.

Notice 2024-04 Section 3 § 1.9 expects the AIS program to "include processes and procedures providing notice to impacted consumers that AI systems are in use and provide access to appropriate levels of information based on the phase of the insurance life cycle in which the AI systems are being used." Two duties sit inside one sentence: a disclosure duty (tell the affected consumer AI is in the loop) and a graduated access duty (give information calibrated to the life-cycle phase — a marketing-stage touch and an adverse claim determination do not warrant the same disclosure depth). The notice does not prescribe wording, timing or channel, and does not create a standalone right of access; it is an expectation about the insurer having a documented process. Transparency and explainability to the impacted consumer are separately one of the five proportionality factors in Section 3 that set how heavy the surrounding controls must be, so a deployment with low explainability raises the required control intensity elsewhere in the programme rather than being prohibited.