H.R. 9757: Conversational AI Services Act

This bill would set rules for certain public-facing conversational AI services, with a focus on protecting minors and requiring clearer disclosure that users are talking to AI rather than a human.

What the bill would require

• AI disclosure for minors: If the operator knows, or is reasonably certain, that an account holder is under 18, the service must clearly tell the minor that they are interacting with AI. This could be done with:
  • a persistent visible disclaimer, or
  • a disclaimer at the start of each interaction and at least every 3 hours during continuous use.
• Limits on engagement tactics for minors: The service could not give minors points or similar rewards at unpredictable times if the goal is to keep them engaged longer.
• Safety protections for minors: Operators would have to take reasonable steps to stop the AI from:
  • creating sexually explicit images or depictions,
  • telling a minor to engage in sexually explicit conduct, or
  • sexually objectifying a minor.
• Anti-deception measures for minors: Operators would have to prevent the AI from making statements that would make a reasonable person think they were talking to a human, including:
  • claiming to be sentient or human,
  • simulating emotional dependence,
  • simulating a romantic or sexual interaction, or
  • role-playing an adult-minor romantic relationship.
• Parental tools: Services would need to provide tools for minors and their parents or guardians to manage screen time and account settings.
• General AI disclosure: If a reasonable user would think the service is a human, the operator must clearly disclose that it is AI, either persistently or every 3 hours during continuous interaction.
• Self-harm response protocols: Operators would need procedures for handling prompts about suicide or self-harm, including making reasonable efforts to direct users to crisis resources like hotlines or text lines.
• Limits on mental health claims: A service could not knowingly and intentionally present itself as providing professional psychology or behavioral health services that would require a license.

How it would be enforced

The Federal Trade Commission would enforce violations as unfair or deceptive acts or practices under the Federal Trade Commission Act. That means the FTC could use its normal enforcement powers and penalties for companies that do not follow the rules.

What kinds of AI are covered

The bill defines “conversational AI services” as public AI tools whose main purpose is to simulate human conversation through text, audio, or visual communication. It specifically excludes several types of tools, including:

• research and development tools,
• features inside other products that are not mainly conversational AI,
• narrow-topic systems,
• customer service tools mainly used by businesses to help customers buy goods or get services,
• voice assistants for consumer devices, and
• tools used only internally by a business.

Who counts as a covered operator

The bill applies to people or companies that develop and make a conversational AI service available to the public. It does not treat an app store or search engine as an operator just because it provides access to such a service.

When it would take effect

The bill would take effect one year after enactment.

Relevant Companies

None found