Pennsylvania HB 2800 (Artificial Intelligence Risk Prevention Act)

House Bill 2800 of the Pennsylvania General Assembly's 2025–2026 regular session, cited in its text as the Artificial Intelligence Risk Prevention Act, is a bill on catastrophic risk from frontier AI models. It was introduced on September 22, 2026 by Rep. Melissa Shusterman (D-Chester) with 14 Democratic co-sponsors, printed as Printer's Number 3904, and referred to the House Communications and Technology Committee on September 23, 2026 (Source: palegis.us) (Source: palegis.us). Its long title describes it as providing for "artificial intelligence risk prevention; establishing standards for frontier developers in addressing critical safety incidents and catastrophic risks; imposing duties on the Pennsylvania Emergency Management Agency and the Attorney General; and imposing penalties" (Source: palegis.us).

The sponsors state that the bill is "based on language from laws already passed with bipartisan support in other states, including California, Illinois and New York" (Source: pahouse.com). Its definitions and thresholds match those of California SB 53, the Transparency in Frontier Artificial Intelligence Act; it also provides for an annual third-party audit, a registration and ownership-disclosure regime, and enforcement by the state emergency-management agency.

Sources for the table: the General Assembly bill page and the printed bill (Source: palegis.us) (Source: palegis.us).

Status and timeline

Shusterman and Rep. Greg Scott (D-Montgomery) circulated a co-sponsorship memorandum, "Addressing Catastrophic Risks from Frontier Artificial Intelligence Models", in August 2026, and announced the forthcoming bill on September 14, 2026 (Source: pahouse.com) (Source: palegis.us). The announcement cited news coverage "over the past week" of concerns expressed by chief executives of major AI companies about risks from the largest models, including "catastrophic failures of the worldwide internet, bioweapons development, threats to financial, utility and infrastructure systems", and stated that "in the absence of action at the federal level" Pennsylvania should act (Source: pahouse.com).

The bill was introduced on September 22 and referred to committee on September 23, 2026. As of September 25 the General Assembly listed no committee meetings or votes (Source: palegis.us).

Scope and definitions

Section 102 defines a frontier model as a foundation model trained using more than 10^26 integer or floating-point operations, counting the original training run together with any subsequent fine-tuning, reinforcement learning or other material modification the developer applies. A frontier developer is a person who trains or initiates training of such a model, and a large frontier developer is a frontier developer that, with its affiliates, had annual gross revenues above $500,000,000 in the preceding calendar year (Source: palegis.us). These thresholds match those of California SB 53 (California SB 53). The text-layer extraction of the printed bill renders the compute threshold as "1,026", a loss of the superscript; the definition's structure and the SB 53 correspondence indicate 10^26.

Catastrophic risk is a foreseeable and material risk that a frontier developer's development, storage, use or deployment of a frontier model will materially contribute to the death of or serious injury to more than 50 people, or more than $1,000,000,000 in property damage or loss, arising from a single incident in which the model provides expert-level assistance with a chemical, biological, radiological or nuclear weapon; engages without meaningful human oversight in a cyberattack or in conduct that would be murder, assault, extortion or theft if committed by a human; or evades the control of its developer or user. The definition excludes publicly accessible information, lawful activity of the federal government, and harm to which the model did not materially contribute when combined with other software; loss of equity does not count as property damage (Source: palegis.us).

A critical safety incident is unauthorized access to, modification of or exfiltration of a frontier model's weights that results in death or bodily injury; harm from the materialization of a catastrophic risk; loss of control of a frontier model causing death or bodily injury; or a frontier model's use of deceptive techniques against its developer to subvert the developer's controls or monitoring, outside an evaluation designed to elicit that behavior, in a manner that demonstrates materially increased catastrophic risk (Source: palegis.us).

"Deploy" means making a frontier model available to a third party for use, modification, copying or combination with other software, and excludes making it available primarily for research, development or evaluation of the model (Source: palegis.us).

Key provisions

Registration and disclosure (Chapter 3). The Pennsylvania Emergency Management Agency, in consultation with the Attorney General, is to develop a registration form for any person who wishes to operate as a frontier developer in the Commonwealth. From January 1, 2027, a large frontier developer may not develop, deploy or operate a frontier model in Pennsylvania without first registering, filing a current disclosure statement and paying an annual fee. The disclosure statement lists the developer's trade names, its principal place of business and Pennsylvania offices, three points of contact, and, for a privately or closely held developer, every person holding a 5% or greater beneficial interest and every person who held one in the preceding five years; for a publicly traded developer or parent, holders of a 50% or greater interest. It is refiled annually, on transfer of ownership of a frontier model, and on any material change. The agency sets the fee by regulation and publishes a list of filers, omitting the contact details (Source: palegis.us).

Frontier AI frameworks (section 501). Each large frontier developer must develop a frontier AI framework covering ten matters: incorporating national and international standards and industry-consensus practices; thresholds for catastrophic-risk capabilities, which may be tiered; mitigations; review of assessments and mitigations before deployment or extensive internal use; third-party assessment of risk and of mitigation effectiveness; criteria for updating the framework; cybersecurity for unreleased model weights; identifying and responding to critical safety incidents; internal governance; and catastrophic risk from internal use, including a model circumventing oversight mechanisms. Implementation begins January 1, 2028; the framework must be published, complied with, reviewed at least annually, and republished with a justification within 30 days of any material modification (Source: palegis.us).

Transparency reports (section 502). Before or concurrently with deploying a new or substantially modified frontier model, every frontier developer must publish a report giving its website, a contact mechanism, the release date, supported languages and output modalities, intended uses, and general use restrictions. A large frontier developer must add summaries of each catastrophic-risk assessment under its framework, the results, the extent of third-party evaluator involvement, and other steps taken under the framework, in machine-readable form. Publication within a system card or model card satisfies the requirement (Source: palegis.us).

Independent audits (Chapter 7). From January 1, 2028, or 90 days after a developer first qualifies as a large frontier developer if later, each large frontier developer must annually retain a third party to audit its compliance with the act. The auditor must follow generally accepted auditing standards and have demonstrated competence, including staff with technical expertise in frontier-model safety; neither party may hold a financial interest in the other, and payment may not depend on the audit's results. Auditors receive access to all reasonably necessary materials, including unredacted versions of published documents, subject to security protocols such as on-premises review. The audit report covers substantial compliance, any material deviation, a detailed assessment of internal controls, the audit personnel, conflict-of-interest procedures and methodology, and is signed by the lead auditor. Within 30 days the developer must publish a high-level summary and a redacted copy and transmit the redacted report to the agency and the Attorney General, and must retain the unredacted report for as long as the model is deployed plus five years (Source: palegis.us). The sponsors described this as "requiring third parties to conduct independent audits for compliance" (Source: pahouse.com).

Critical safety incidents (Chapter 9). The agency is to establish a reporting mechanism open to frontier developers and to the public. A frontier developer must report a critical safety incident to the agency and the Attorney General within 72 hours of learning facts sufficient for a reasonable belief that one has occurred, and within 24 hours to an appropriate law-enforcement or public-safety authority where the incident poses an imminent risk of death or serious physical injury. The agency and the Attorney General may transmit reports to the federal government, the Governor, state agencies and the General Assembly, and must restrict access to need-to-know personnel (Source: palegis.us).

Internal-use risk and misstatements (Chapter 11). A large frontier developer must confidentially transmit summaries of its assessments of catastrophic risk from internal use of its models every three months or on another agreed schedule. No frontier developer may make a materially false or misleading statement about catastrophic risk from its models or its management of it, and no large frontier developer may do so about its implementation of or compliance with its framework, with an exception for statements made in good faith and reasonable in the circumstances (Source: palegis.us).

Whistleblower protections (section 1302). An employee responsible for assessing, managing or addressing critical-safety-incident risk may disclose, to the agency, the Attorney General, a federal authority or an internal authority, information the employee has reasonable cause to believe shows a specific and substantial danger to public health or safety from a catastrophic risk, or a violation of the act. Large frontier developers must maintain an anonymous internal reporting channel with monthly status updates to the reporter and quarterly reporting to officers and directors. Developers may not restrict such disclosures by policy or contract or retaliate against them, and must give annual or posted notice of the rights. The section preserves the Pennsylvania Whistleblower Law of 1986 (Source: palegis.us).

Enforcement and penalties

The Pennsylvania Emergency Management Agency enforces the act in consultation with the Attorney General (section 1301). A person who develops, deploys or operates a frontier model in Pennsylvania without meeting the act's requirements, or who submits materially false information, faces, after notice and hearing, a civil penalty of up to $1,000 per day of violation plus the agency's costs. A large frontier developer that fails to publish or transmit a required document, fails to comply with its own framework, fails to obtain an audit, fails to report a critical safety incident, or makes a prohibited misstatement faces a civil penalty scaled to severity of up to $1,000,000 per violation for a first violation and $3,000,000 per violation for a subsequent one. Civil penalties are recovered only in an action brought by the Attorney General and deposited in the General Fund; the agency may also seek injunctive relief. The bill expressly creates no private right of action (Source: palegis.us).

Audit reports, auditors' working papers, incident reports, internal-use risk assessments and employee reports held by the agency or the Attorney General are exempt from the Right-to-Know Law. From January 1, 2029 the agency is to report annually to the Governor and the General Assembly with anonymized, aggregated incident information and any recommended amendments. The agency must issue temporary regulations within six months of the effective date of section 1504; the remainder of the act takes effect one year after notice of those temporary regulations is published in the Pennsylvania Bulletin (Source: palegis.us).

Comparison with other state frontier-AI laws

The sponsors named California, Illinois and New York as the states whose enacted laws the bill draws on (Source: pahouse.com). Against California SB 53, the bill uses the same 10^26-operation and $500 million thresholds, the same catastrophic-risk harm levels (more than 50 deaths or serious injuries, or more than $1 billion in damage), a framework-and-transparency-report structure, incident reporting, a misstatement prohibition and whistleblower protections, and the same $1 million first-violation ceiling (California SB 53). Provisions in the Pennsylvania text beyond that shared core include the annual independent third-party audit, the registration, beneficial-ownership disclosure and fee regime administered by an emergency-management agency, a $3 million ceiling for subsequent violations, and a per-day penalty for operating without meeting the act's requirements (Source: palegis.us). New York's frontier-model law is covered at New York RAISE Act.

Relationships

• instance-of: State-Level AI Regulation
• related: Frontier AI Governance
• related: AI Incident Reporting
• related: AI Whistleblowing
• depends-on: California SB 53
• related: New York RAISE Act
• related: Counting state AI legislation