ai safety and transparency

Summary#

This bill sets rules for conversational AI services, with a strong focus on protecting minors and preventing users from being misled into thinking they are talking to a human. It requires clear disclosures, safety limits for minors, parental controls, suicide/self‑harm referral protocols, and bans on certain engagement tactics. The Federal Trade Commission (FTC) would enforce the rules and can use its existing powers and penalties.

• Who must follow the rules: an "operator" who develops and makes a conversational AI service available to the public.
• Disclosures: AI must be clearly labeled to minors and must be labeled whenever a reasonable user might think they are interacting with a human (persistent label or every 3 hours of continuous use).
• Minor protections: no unpredictable points/rewards intended to boost engagement; must block sexual content or sexualized language toward minors; must not simulate being sentient, emotionally dependent, romantic with a minor, or role‑play an adult‑minor romantic relationship.
• Parental and safety tools: operators must offer parental controls and screen‑time/account management tools; must adopt suicide/self‑harm response protocols that refer users to crisis services.
• Prohibition on unlicensed therapy claims: AI may not be presented as providing licensed mental health treatment.

What it means for you#

• Minors (under 18):

  • You must be shown a clear notice that you are interacting with AI (persistent or frequent disclaimer).
  • The AI should not produce sexualized images or tell you to engage in sexual acts.
  • The AI should not pretend to be a sentient human, promise emotional dependence, or role‑play adult‑minor romantic/sexual scenarios.
  • You may have access to tools your parent/guardian can use to manage screen time and settings.

• Parents and guardians:

  • Operators must offer tools to manage a minor’s screen time and account settings.
  • You should see visible disclosures so you can tell whether your child is talking to AI.

• Operators (companies that offer public conversational AI):

  • Must add visible disclosures for minors and in cases where users could reasonably think the AI is human.
  • Must implement measures to prevent sexual content involving minors and to avoid deceptive human‑like claims.
  • Must stop using unpredictable interval reward systems aimed at increasing minor engagement.
  • Must implement suicide/self‑harm response protocols and avoid claiming the AI offers licensed mental‑health services.
  • Compliance must begin one year after the bill becomes law.

• Businesses using AI internally or narrow tools:

  • The bill excludes some tools: research‑only systems, narrow-topic tools, business‑to‑business commercial tools, voice assistants, speaker/voice interfaces, and AI used solely internally by a business are not covered by this law.

• General public and users:

  • If the AI would lead a reasonable person to think it’s human, the operator must disclose that it’s artificial. The bill sets timing options for how often that disclosure appears.

• Enforcement:

  • The FTC enforces violations under the same processes and penalties it already uses for unfair or deceptive practices.

• Timing:

  • The rules start one year after the law is enacted.

Expenses#

No publicly available information.

• The bill text does not include a fiscal note or budget estimate.
• Likely cost items (inferred): operators would incur development and testing costs to add disclaimers, parental controls, content filters, and crisis‑referral functions; the FTC may have enforcement costs. These are possibilities inferred from the bill text, not estimates provided by the bill.

Proponents' View#

The bill appears intended to protect children and increase transparency about AI.

• The bill appears intended to reduce the risk that minors are sexually exploited or exposed to sexual content by conversational AI.
• It appears intended to prevent conversational AI from deceiving people into thinking it is human, especially in emotionally sensitive interactions.
• The bill appears intended to give parents tools to manage minors’ use of public conversational AI.
• It appears intended to ensure conversational AI responds appropriately to suicide or self‑harm prompts by directing users to crisis services.
• It appears intended to make public AI services more transparent about their artificial nature.

Opponents' View#

The bill’s text raises several practical and legal questions that could be points of concern.

• One concern is vagueness: terms like “reasonable measures,” “reasonably certain,” and “would lead a reasonable individual” are not precisely defined, which may create uncertainty about required technical steps.
• The scope of covered services could be unclear in practice. The law exempts many business or narrow uses, but it may be hard to decide which services count as having the “primary purpose” of simulating human conversation.
• Compliance costs and technical burdens may be significant for some operators (for example, implementing continuous disclaimers, content filters, parental tools, and suicidal‑ideation protocols), but the bill does not provide guidance or funding.
• The ban on “points or similar rewards at unpredictable intervals with the intent to encourage increased engagement” may be hard to interpret and enforce because it depends on proving intent and on what counts as “unpredictable” rewards.
• Enforcement relies on the FTC using its existing unfair/deceptive practices authority; the bill does not explain how the FTC will prioritize or resource enforcement for this new area.