Chatbot Laws · verified Oct. 21, 2026
Would require AI chatbots to disclose they are AI, warn that they are not licensed professionals when users seek help with high-stakes health, estate or financial decisions, detect crises and refer users to help such as 988, and limit use of conversation data. Covered entities would have to curb designs that exploit older adults and report adverse incidents yearly to the FTC and NIH.
Who it would cover
Any person, including an owner or operator, that makes an AI chatbot available to individuals in the United States (covered entity). The duties apply to all users, not only older adults (65 and over). Services limited to predetermined or contextual replies within a narrow purpose are excluded.
What it would require
| AI disclosure | Chatbots must clearly and conspicuously disclose that they are AI and not human, may not falsely claim to be human, and must disclose whether and when the interaction is subject to human oversight or review.When: At the start of each conversation and at reasonably regular intervals during extended interactions§ 3(a)(1)(A)(i)-(ii), (iv) |
|---|---|
| Crisis protocol | Chatbots must use evidence-based methods to spot crisis indicators (suicidal ideation, self-harm, intent to harm others, imminent threats, medical emergencies), refer users to emergency services, a human professional or a crisis line such as 988, prioritize that referral, and not give self-harm, suicide or medication-change advice or encouragement. Covered entities must publish their protocols.§ 2(3); § 3(a)(1)(D)-(E), (a)(2) |
| Reporting | Within 1 year of enactment and yearly after, covered entities report material adverse incidents involving a high-stakes decision or crisis to the FTC and NIH, with counts by age band where known, crisis-response outcomes, detection methods and corrective actions; the FTC and NIH publish de-identified aggregate data at least yearly.§ 4 |
| Design limits | Covered entities must take reasonable steps to keep chatbots from encouraging excessive reliance, exploiting age-related cognitive limits or social isolation, discouraging help from family or professionals, putting chatbot use ahead of real-world relationships, or pressuring or guilting users who try to stop.§ 3(c)(1) |
| Therapy and licensed titles | Chatbots may not falsely claim to be a licensed professional. When they detect a user seeking tailored guidance on a high-stakes decision (health care, estate planning, guardianship, or money and account information), they must say they are not a licensed professional, should not be the sole basis for the decision, and are not covered by professional privileges.§ 2(7); § 3(a)(1)(A)(iii), (B)-(C) |
| Data limits | Conversation data may be collected, kept or disclosed only as needed to provide the service, comply with the Act or other law, prepare reports, or guard against fraud, security incidents or imminent harm; other uses outside covered conversations need affirmative consent, as does training on user information. Users must be able to delete their conversation history.§ 3(a)(3), (b) |
| Other duties | NIST would develop voluntary guidelines on chatbots interacting with older adults (65 and over) within 18 months of enactment.§ 2(9); § 6 |
Enforcement
| Enforced by | Federal Trade Commission and State attorneys general (parens patriae, § 3 only)§ 5 |
|---|---|
| Private lawsuits | No |
| Penalties | Violations of §§ 3-4 are treated as violations of an FTC rule on unfair or deceptive practices. Civil penalty of up to $50,000 for each knowing or reckless violation of § 3, in addition to other penalties. States may seek injunctions, damages, penalties and restitution. |
| Cure period | For a first violation of § 3 or § 4, 60 days after written FTC notice; no civil penalty if cured and certified in writing. Not available for intentional deception, substantial risk of imminent harm, or repeated or ongoing violations. |
Sources
- S. 5117 as introduced (text), GPO govinfo Supports: Titles, sponsor and cosponsor, §§ 1-9.
- Bill status data (Library of Congress feed, GPO govinfo) Supports: Introduction and referral Jul 23, 2026; sponsor and original cosponsor; latest action.