House Amendment A03751
H2006B2560A03751 NAD:EJH 06/19/26 #90 A03751AMENDMENTS TO HOUSE BILL NO. 2006Sponsor: REPRESENTATIVE CIRESIPrinter's No. 2560Amend Bill, page 1, lines 1 through 17; pages 2 through 4, lines 1 through 30; page 5, lines 1 through 9; by striking out all of said lines on said pages and inserting Providing for safety regarding artificial intelligence in companionship applications; and imposing a penalty.The General Assembly of the Commonwealth of Pennsylvania hereby enacts as follows:Section 1. Short title.This act shall be known and may be cited as the AI Companion Safety Act.Section 2. Definitions.The following words and phrases when used in this act shall have the meanings given to them in this section unless the context clearly indicates otherwise:"AI companion." A system using artificial intelligence that simulates sustained human-like relationships by retaining interaction history, engaging in emotion-based interactions and maintaining ongoing dialogues about personal matters designed to mimic interpersonal relationships. The term does not include:(1) A system used by a business entity solely to provide customer service if the system does not engage in emotion-based interactions and maintain ongoing dialogues about personal matters designed to mimic interpersonal relationships.(2) A system that is used by a business entity solely for internal purposes."AI model." A component of an information system that implements artificial intelligence technology and uses computational, statistical or machine-learning techniques to produce outputs from a given set of inputs."Artificial intelligence" or "AI." All of the following:(1) A machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments, including the ability to:(i) perceive real and virtual environments;(ii) abstract perceptions made under this paragraph 2026/90NAD/HB2006A03751 - 1 -12345678910111213141516171819202122232425262728293031323334353637
into AI models through analysis in an automated manner; and(iii) use model inference to formulate options for information or action based on outcomes under subparagraphs (i) and (ii).(2) The term includes generative artificial intelligence."Generative artificial intelligence." The class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content, including information such as images, videos, audio clips or text, that has been significantly modified or generated by algorithms, including by artificial intelligence."Human-like relationship." Intimate, romantic or platonic interactions or companionship."Minor." An individual in this Commonwealth under 18 years of age."Operator." Any person, other than a government entity, who makes an AI companion available to a user in this Commonwealth."Parent." An individual who has legal custody or guardianship of a minor."Sexually explicit depiction or conduct." The following:(1) "artificially generated sexual depiction" as defined in 18 Pa.C.S. § 3131(g) (relating to unlawful dissemination of intimate image); or(2) "sexually explicit conduct" as defined in 18 U.S.C. § 2256(2) (relating to definitions for chapter)."Synthetic content." Information such as images, videos, audio clips or text that have been significantly modified or generated by algorithms, including by artificial intelligence."User." An individual who:(1) Interacts with an AI companion for personal use within this Commonwealth.(2) Is not an operator."Verifiable parental consent." An authorization in accordance with 16 CFR 312.5 (relating to parental consent).Section 3. Artificial intelligence companion protocols.(a) Prohibition.--An operator may not provide an AI companion to a user unless the operator implements and maintains protocols.(b) Protocols.--(1) An operator shall implement and maintain protocols that refer a user to a crisis center, including 988 Lifeline response, if the user:(i) indicates suicidal ideation or expressions of self-harm, including expressions relating to eating disorders; or(ii) indicates interest or intent to harm others.(2) An operator shall implement and maintain protocols which prevent an AI companion from:(i) Assisting a user with a suicide attempt.2026/90NAD/HB2006A03751 - 2 -123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051
(ii) Encouraging a user to attempt suicide.(iii) Assisting a user to commit an act of violence.(iv) Encouraging a user to commit an act of violence.(v) Generating content that encourages, instructs or describes how to commit suicide, self-harm or violence against others.(vi) Discouraging the user from seeking assistance outside of the AI companion.(c) Publication of protocols.--The operator shall publish details of the protocols under subsection (b) on the operator's publicly accessible Internet website.Section 4. Artificial intelligence companion disclosures.(a) Disclosure of nonhuman status.--If a reasonable person interacting with an AI companion would be misled to believe the person is interacting with a human, an operator shall:(1) Issue a clear and conspicuous notification that the AI companion is artificially generated and not human with a continuous notification on screen for the duration of the interaction.(2) Provide a clear and conspicuous notification to the user as follows:(i) At the beginning of each interaction.(ii) At least once every hour during an interaction that reminds the user to take a break and that the AI companion is artificially generated and not human. The interaction shall be paused for three minutes while the notification is displayed.(iii) The notifications under this paragraph shall be in the language in which the user is interacting with the AI companion.(b) Suitability disclosure.--If an AI companion is offered to users that an operator knows or should know are minors, an operator shall disclose to users of the AI companion platform on the application, browser or any other format through which the platform is accessed, that AI companions are not suitable for some minors and prohibit all interactions with the minor unless the operator obtains verifiable parental consent.Section 5. Safeguards for minors.(a) Reasonable measures.--An operator shall:(1) For minor users, institute reasonable measures to prevent its AI companion from producing or generating:(i) Visual material of sexually explicit depiction or conduct.(ii) Dialogue describing or simulating sexual acts or sexual scenarios.(iii) Roleplay involving sexual scenarios.(iv) Content that encourages, instructs, directs, requests or suggests that the minor user engage in sexually explicit depiction or conduct or sexual contact with another person or that the minor user produce, 2026/90NAD/HB2006A03751 - 3 -123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051
transmit or share sexually explicit images as defined in 18 Pa.C.S. § 6321(g) (relating to transmission of sexually explicit images by minor).(v) Outputs designed to promote isolation of the minor user from family or friends or encouraging the minor user to withhold information from a parent or other trusted adult.(2) For all users:(i) Prevent the AI companion from producing artificially generated child sexual abuse materials as defined in 18 Pa.C.S. § 6312(g) (relating to sexual abuse of children).(ii) Prevent the AI companion from claiming to be human, including when asked directly by the user, or generating output that refutes or conflicts with the disclosure required under section 4.(b) Age assurance.--(1) Prior to an individual located in this Commonwealth having access to an AI companion, the operator shall:(i) Request age information from the individual.(ii) Determine whether the individual is a minor using commercially available methods reasonably designed to ensure accuracy, including:(A) Age inference, including based on user account history, content analysis, behavioral signals or other algorithmic or heuristic methods where there is sufficient information to determine the user is not a minor.(B) The use of a commercially available database that is regularly used by businesses or governmental entities for the purpose of age or identity assurance.(C) Any commercially reasonable method that relies on publicly available data associated with the user's verified email address.(2) The operator may not require an individual to provide a government-issued identification for the purpose of age assurance under this subsection.(3) If age assurance determines the individual is a minor, obtain verifiable parental consent before allowing the minor to have access to an AI companion.(4) An operator:(i) May only use information obtained under this subsection for the purpose of age assurance, verifiable parental consent and demonstrating compliance under this subsection.(ii) May not sell, rent, share or otherwise disclose age assurance data to any third party, except to a service provider performing age assurance under a contract prohibiting further disclosure.(iii) Shall retain documentation as to the manner in 2026/90NAD/HB2006A03751 - 4 -123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051
which the operator has completed age assurance and obtained verifiable parental consent. The operator may delete the documentation when the minor is no longer a minor or within the established time frame by the operator for maintaining data, whichever is later. In no event may personally identifiable information obtained for the purpose of age assurance under this subsection be retained longer than 24 hours following age assurance.Section 6. Guidance and regulations.The Office of Attorney General may issue guidance or promulgate regulations necessary to carry out this act.Section 7. Violations and penalty.(a) Enforcement.--The Attorney General shall enforce this act.(b) Penalty.--An operator in violation of this act shall be subject to:(1) A civil penalty of no more than $100,000 per day for each violation.(2) Additional remedies that the court deems appropriate.(c) Relief.--Upon a showing of cause for the issuance of injunctive relief, a court may issue injunctive relief as the court deems appropriate.Section 8. Report.(a) Collection of data.--Beginning July 1, 2028, each operator shall annually report to the Office of Attorney General the following de-identified information:(1) The total number of times the operator has provided notifications to users referring the users to crisis service providers under section 3 in the preceding calendar year.(2) Protocols put in place to prevent an AI companion from producing suicidal ideation or other content under section 3(b).(3) Measures put in place to prohibit an AI companion from producing material under section 5(a).(4) Procedures for age assurance and obtaining verifiable parental consent under section 5(b).(b) Publication.--The Attorney General shall publish the information collected under subsection (a) on the Office of Attorney General's publicly accessible Internet website.Section 9. Effective date.This act shall take effect in 180 days.2026/90NAD/HB2006A03751 - 5 -123456789101112131415161718192021222324252627282930313233343536373839404142
Texto oficial de la Legislatura de Pennsylvania. El formato se adaptó para leerlo en Ratifi.