HB 1898

This bill prohibits a frontier model developer, excluding an accredited college or university to the extent that the college or university is developing or using frontier models exclusively for academic research purposes, or a person who makes a covered chatbot, which is a service that allows an ordinary person to have conversations whe re humanlike responses are generated by a foundation model, is foreseeably likely to be accessed by minors, and has at least 1 million monthly active users, available in this state and who, together with the person's affiliates, collectively had an annual r evenue of at least $25 million ("large chatbot provider"") from making a materially false or misleading statement about a catastrophic risk or a child safety risk (together, ""covered risk"") from the frontier developer or large chatbot provider's activities or their management of covered risks. Further, a frontier developer who together with its affiliates collectively had annual revenue of at least $500 million (""large frontier developer"") and a large chatbot provider is prohibited from making a materiall y false or misleading statement about its implementation of, or compliance with, its public safety plan or child safety plan. Large Frontier Developers This bill requires a large frontier developer to implement and clearly publish on its website a public safety plan that describes in detail how the large frontier developer defines and assesses thresholds used by the developer to determine whether a fron tier model has capabilities that could pose a catastrophic risk. As used in this bill, a ""catastrophic risk"" means a foreseeable and material risk that a frontier developer's development, storage, use, or deployment of a frontier model will materially co nt ribute to the death of, or serious injury to, 50 people or more than $1 billion in damage to, or loss of, property arising from a single incident involving (i) providing expert-level assistance in the creation or release of a chemical, biological, radiolo gical, or nuclear weapon; (ii) engaging in conduct with no meaningful human oversight, intervention, or supervision that is either a cyberattack or, if the conduct had been committed by a human, would constitute the crime of murder, assault, extortion, or t heft, including theft by false pretense; or (iii) evading the control of its frontier developer or user. However, catastrophic risk does not include a foreseeable and material risk from (i) information that a frontier model outputs if the information is otherwise publicly accessible in a similar form; (ii) lawful activity of the federal government; or (iii) harm caused by a frontier model in combination with other software if the frontier model did not materially contribute to the harm. This bill requires the plan to also describe how the large frontier developer addressed all of the following:  Applies mitigations to address the potential for catastrophic risks.  Reviews assessments of catastrophic risk as part of the decision to deploy a frontier model or use it internally.  Uses third parties to assess the potential for catastrophic risks and the effectiveness of mitigations of catastrophic risks.  Implements cybersecurity practices to secure unreleased frontier model weights, which are numerical parameter in a frontier model that is adjusted through training and that helps determine how inputs are transformed into outputs, from unauthorized modification or transfer by internal or external parties.  Assesses and manages catastrophic risk resulting from the internal use of the frontier developer's frontier models.  Incorporates national standards, international standards, and industry-consensus best practices into the large frontier developer's public safety plan.  Revisits and updates the public safety plan.  Identifies and responds to ""critical safety incidents, which means (i) unauthorized access to, or modification, inadvertent release, or exfiltration of, the model weights of a frontier model; (ii) the death of, or serious injury to, more than 50 people or more than $1 billion dollars in damage to, or loss of, property resulting from the materialization of a catastrophic risk; (iii) loss of control of a frontier model that cause death or bodily injury, or that demonstrates materially increased catastrophic risk; or (iv) a frontier model that uses deceptive techniques against the frontier developer to subvert the controls or monitoring of its frontier developer outside of the context of an evaluation designed to elicit such behavior and in a manner that demonstrates materially increased catastrophic risk. Loss of value of equity does not count as damage to or loss of property.  Institutes internal governance practices to ensure implementation of the public safety plan. This bill requires a large frontier developer to clearly publish any material modifications to its public safety plan, along with a justification for the modification, within 30 days of making the material change. Before a large frontier developer deploys a new frontier model, this bill requires the large frontier developer to publish summaries of any assessments of catastrophic risks from the frontier model, the results of the assessments, the extent to which thi rd-party evaluators were involved in the assessments, and other steps taken to fulfill the requirements of the public safety plan. This information may be published as part of a larger document, including a system card or model card. Large Chatbot Providers This bill requires a large chatbot provider to implement and clearly publish on its website a child safety plan that describes in detail how the large chatbot provider assesses potential for child safety risks. As used in this bill, a ""child safety risk"" means a material and foreseeable risk that a frontier developer's artificial intelligence model that is trained on a broad data set, designed for generality of output, and adaptable to a wide range of distinctive tasks (""foundation model""), when used as part of a covered chatbot operated by the frontier developer, will engage in behavior when interacting with a minor that, if it had been engaged in by a human, would be d eemed to intentionally or recklessly cause death or bodily injury to the minor, including as a result of self-harm; or damage to the mental health of such minor that constitutes severe emotional distress. This bill requires the plan to also describe how the large chatbot provider addresses all of the following:  Applies mitigations to address the potential for child safety risks.  Uses third parties to assess the potential for child safety risks and the effectiveness of mitigations of child safety risks.  Incorporates national standards, international standards, and industry-consensus best practices into the large chatbot provider's child safety plan.  Revisits and updates the child safety plan.  Identifies and responds to ""child safety incidents, which means a covered chatbot engaging in behavior when interacting with a minor that, if the behavior had been engaged in by a human, would be deemed to intentionally or recklessly cause death or bodily injury to such minor or damage to the mental health of such minor that constitutes severe emotional distress.  Institutes internal governance practices to ensure implementation of the child safety plan. This bill requires a large chatbot provider to clearly publish any material modifications to its child safety plan, along with a justification for the modification, within 30 days of making the material change. Before a large chatbot provider integrates a foundation model into a covered chatbot, this bill requires the large chatbot provider to publish summaries of any assessments of child safety risks, the results of the assessments, the extent to which third-p arty evaluators were involved in the assessments, and steps taken to fulfill the requirements of the large chatbot provider's child safety plan. Confidentiality This bill authorizes a large frontier developer or chatbot provider to make redactions to published safety plan documents, if the redactions are necessary to protect the large frontier developer or large chatbot provider's trade secrets or cybersecurity, public safety, or the national security of the United States, or to comply with federal or state law. However, the developer or chatbot provider must describe the character and justification of the redactions and retain the redacted information for at l ea st five years. REPORTING OF INCIDENTS This bill requires the attorney general to establish a means for reporting a child safety incident or a critical safety incident (together, ""safety incident""). The form must allow the report to include, at least, the date of the safety incident, the rea sons the incident qualifies as a safety incident, and a short and plain statement describing the safety incident. A frontier reporter is required to report a critical safety incident to the attorney general within 15 days of discovery. If a critical saf et y incident poses an imminent risk of death or serious physical injury, then this bill requires the frontier developer to disclose the incident within 24 hours to an appropriate authority. This bill requires a large chatbot provider to report a child safe ty incident to the attorney general within 15 days of discovery. This bill requires the attorney general to establish a mechanism for a large frontier developer to confidentially submit summaries of any assessments of catastrophic risks resulting from internal use of frontier models. Large frontier developers must tr ansmit to the attorney general a summary of such assessment beginning on January 1, 2027, and every three months thereafter. This bill authorizes the attorney general to transmit reports of safety incidents, summaries of assessments of catastrophic risk resulting from internal use, and reports from employees to the general assembly, governor, federal government, or appropriate state agencies. In transmitting such reports, the attorney general may consider any risks related to trade secrets, public safety, cybersecurity, or national security. This bill requires the department of safety and attorney general to designate one or more federal laws or guidance documents that impose standards or requirements for safety incident reporting that are equivalent to or stricter than the reporting requirements described above and is intended to assess, detect, or mitigate catastrophic or child safety r isk. If a frontier developer or large chatbot provider intends to comply with reporting requirements of such designated federal law or guidance, then it must declare its intent to do so to the attorney general and department of safety. After declaring s h intention, the frontier developer or large chatbot provider is in compliance with this bill to the extent that it meets the requirements of the designated federal law or guidance. However, the failure of a frontier developer or large chatbot provider t o comply with the designated federal law or guidance constitutes a violation of this bill. CIVIL PENALTIES This bill provides that a large frontier developer that violates this bill is subject to a civil penalty of $1 million or less per violation for a first violation and $3 million or less for subsequent violations. On the other hand, a large chatbot provi der that violates this bill is subject to a civil penalty of $50, 000 or less per violation. The attorney general has the exclusive right to enforce this bill. RULEMAKING This bill authorizes the department of safety, in consultation with the office of the attorney general, to promulgate rules to effectuate this bill. APPLICABILITY This bill applies to conduct occurring on or after January 1, 2027. ON APRIL 16, 2026, THE HOUSE ADOPTED AMENDMENTS #1, 2, AND 3, AND PASSED HOUSE BILL 1898, AS AMENDED. AMENDMENT #1 makes the following changes:  Clarifies that ""covered chatbot"" does not include (i) software that operates solely for the purpose of customer service, business operations, productivity and analysis related to source information, internal research, or technical assistance, regardless of the software's capability to use natural language inputs and generate natural language outputs; (ii) software that is incorporated into a video game and cannot discuss topics related to mental health, self-harm, or sexually explicit content, or maintain a dialogue on other topics unrelated to the video game; or (iii) software that operates solely in connection with a specific physical location, such as a theme park or location-based experience, at physical events, or for the promotion, marketing, or exhibition of a film, television program, or other audiovisual work, and which cannot maintain a dialogue on topics unrelated to the physical location, event, or such work.  Provides that ""frontier developer"" means a person who has used, or initiated the use of, a quantity of computing power at least 10^26 integer or floating-point operations to train a frontier model, including computing used for the original training run and for any subsequent fine-tuning, reinforcement learning, or other modifications the person applies.  Provides that ""monthly active users"" means the total global number of unique users who have directly interacted with a service at least once in the preceding 30 days, whether or not those users have registered accounts with the service.  Provides that a large chatbot provider is not required to make a disclosure when integrating a modified version of an existing foundation model into its covered chatbot if (i) the modification to the foundation model was made by a person unaffiliated with the large chatbot provider; (ii) the modification does not increase child safety risks from the covered chatbot; or (iii) the modification was made automatically as part of a continuous learning process and the large chatbot provider has made a disclosure with respect to the foundation model in the preceding 90 days. AMENDMENT #2 makes the following changes:  Adds the Tennessee bureau of investigation and the department of finance and administration for who the department of safety is required to consult with to promulgate rules.  Changes the effective date from January 1, 2027, to July 1, 2027. AMENDMENT #3 rewrites the bill to, instead, require a large chatbot provider to write, implement, comply with, and clearly and conspicuously publish on its internet website a child safety plan that describes in detail how the large chatbot provider does a ll of the following:  A ssess potential for child safety risks .  Applies mitigations to address the potential for child safety risks based on the results of the assessments undertaken pursuant to this amendment.  Uses third parties to assess the potential for child safety risks and the effectiveness of mitigations of child safety risks.  Revisits and updates the child safety plan, including any criteria that triggers updates and how the large chatbot provider determines when its foundation models are substantially modified enough to require disclosures.  Identifies and responds to child safety incidents.  Institutes internal governance practices to ensure implementation of the child safety plan. As used in this amendment, a ""child safety risk"" means a material and foreseeable risk that a frontier developer's foundation model, when used as part of a covered chatbot, will engaged in behavior when interacting with a minor that, if it had been engag ed in by a human, would be deemed to intentionally or recklessly cause death or bodily injury to the minor or damage to the mental health of such minor that constitutes severe emotional distress. Also, a ""child safety plan"" means a documented technical a nd organizational protocol to manage, assess, and mitigate child safety risks. REQUIRED DISCLOSURES If a large chatbot provider makes a material modification to the large chatbot provider's child safety plan, then this amendment requires the large chatbot provider to publish the modified child safety plan and a justification for such modification withi n 30 days of making such material change. As used in this amendment, a ""large chatbot provider"" means a person who makes a covered chatbot available in this state and who, together with the person's affiliates, collectively had an annual revenue of at le as t $25 million. Also, a ""covered chatbot"" means a service that allows an ordinary person to have conversations where humanlike responses are generated by a foundation model that is likely to be accessed by minors and has at least one million monthly activ e users. Before integrating a new foundation model, or substantially modifying an existing foundation model, this amendment requires a large chatbot provider to publish summaries of (i) assessments of child safety risks conducted pursuant to the large chatbot pro vider's child safety plan; (ii) the results of such assessments; (iii) the extent to which third-party evaluators were involved in the assessments; and (iv) other steps taken to fulfill the requirements of the child safety plan. However, such a disclosur e is not required when integrating a modified version of an existing foundation model into its covered chatbot if (i) the modification was made by a person unaffiliated with the large chatbot provider; (ii) the modification does not increase the child safet y risks from the covered chatbot; or (iii) the modification was made automatically as part of a continuous learning process and the large chatbot provider has made a disclosure with respect to the foundation model in the last 90 days. As used in this ame nd ment, a ""foundation model"" means an artificial intelligence model that is trained on a broad data set, designed for generality of output, and adaptable to a wide range of distinctive tasks. This amendment prohibits a large chatbot provider from making a materially false or misleading statement or omission about child safety risks from the large chatbot provider's activities or the large chatbot provider's management of child safety risks. Further, a large chatbot provider is prohibited from making a materially false or misleading statement or omission about its implementation of, or compliance with, a child safety plan. These prohibitions do not apply to statements made in good faith that a re reasonable under the circumstances. This amendment authorizes a large chatbot provider to make redactions to documents required to be published by this amendment. Such redactions must be necessary to protect the large chatbot provider's trade secrets or cybersecurity, or public safety, th e national security of the United States, or to comply with federal or state law. However, the large chatbot provider must describe the character and justification of the redaction and retain the unredacted information for at least five years. CHILD SAFETY INCIDENTS This amendment requires the attorney general to establish a process for a large chatbot provider or member of the public to report a child safety incident. The report must include (i) the date of the child safety incident; (ii) the reasons the incident qualifies as a child safety incident; and (iii) a short and plain statement describing the child safety incident. A large chatbot provider must report any child safety incident within 15 days of discovery. Such a report is not open to public inspection. A s used in this amendment, a ""child safety incident"" means a covered chatbot engaging in behavior when interacting with a minor that, if the behavior had been engaged in by a human, would be deemed to intentionally or recklessly cause death or bodily injur y to such minor or damage to the mental health of such minor that constitutes severe emotional distress. The attorney general must transmit reports of child safety incidents and reports from employees to the general assembly, governor, federal government, or appropriate state agency. Further, the attorney general may consider any risks related to trade secrets, public safety, cybersecurity, or national security when transmitting such reports. This amendment requires the department of safety, in consultation with the attorney general, Tennessee bureau of investigation, and the department of finance and administration, to promulgate rules designating one or more federal laws or guidance documents that (i) impo se or state standards for child safety incident reporting that are substantially equivalent to or stricter than those required by this amendment for child safety incidents; and (ii) are intended to assess, detect, or mitigate child safety risks. Such doc um ent does not need to require child safety incident reporting to this state. A large chatbot provider may comply with this amendment by complying with the requirements of such identified federal law or guidance document, but must declare its intent to do so to the attorney general and department of safety. A large chatbot provider is in compliance with this amendment if it compiles with the identified federal law or guidance document, until it revokes such a declaration or until the department of safety re moves the applicable federal law or guidance document from the rules. The failure to comply with the identified federal law or guidance document constitutes a violation of this amendment if the large chatbot provider is not otherwise in compliance with t his amendment. PENALTIES This amendment provides that a large chatbot provider that violates this amendment is subject to a civil penalty of up to $50, 000 per violation. The attorney general has the exclusive authority to enforce this amendment. CONFLICTS This amendment provides that the duties and obligations imposed by it are cumulative with any other duties or obligations imposed under another law. RULE PROMULGATION This amendment authorizes the department of safety to promulgate rules necessary to effectuate this amendment."