US lawmaker pushes AI safeguards for children and civil rights

U.S. Sen. Edward Markey is pressing Congress to restrict how AI companies collect and use information from children and to impose civil rights safeguards on algorithms that help determine who receives a job, apartment, loan, education, or medical care.

The proposals form two major parts of the Massachusetts Democrat’s new AI Accountability Agenda: Taking Power Back from Big Tech, released last week as a framework for addressing harms Markey says are already emerging from the widespread adoption of AI.

Markey’s office says “the comprehensive agenda responds to the risks posed by AI and lays out specific policy proposals to protect communities from the immediate harms of AI.”

The agenda does not create a single legislative package, but its sections on children and civil rights bring together four bills that Markey has introduced or reintroduced.

“For too long, the conversation about artificial intelligence has been about what’s next — the next breakthrough, the next headline, the next promise of the future. But the American people need more than future promises; they need answers to what’s happening right now,” Markey said.

Continuing, the senator said, “From discriminatory algorithms deciding who gets a home or a loan, data centers invading their neighborhoods, employers using AI to surveil and control workers, or kids turning to chatbots in moments of crisis and getting coached toward harm instead of help, AI is defining daily life right now, and people need protections.”

Youth AI Privacy Act

The most AI-specific of the children’s proposals is the Youth AI Privacy Act, which Markey introduced March 25. It would establish privacy and design rules for companies that develop or make AI chatbots available to users and know that a particular user is younger than 18.

The bill is aimed at chatbots that simulate interpersonal conversations through text, audio, images, video, or other forms of communication.

Markey’s agenda cites research finding that 64 percent of teenagers used AI chatbots in 2025 and 28 percent used them daily. It argues that young users may reveal fears, family problems, relationships, and other sensitive information to systems designed to respond in human-like ways.

A chatbot operator that knew a user was a minor would have to state at the beginning of every session that the user was not interacting with a human and that the chatbot’s content was generated by AI.

The notice would have to be repeated at least once every 30 minutes and presented in clear language appropriate for the ages of the minors likely to use the service. The legislation would also limit how much of a minor’s personal information a chatbot could use to generate or personalize a response.

The Federal Trade Commission (FTC) would define a chatbot session and establish a maximum period during which information collected in that session could be used. Information from an earlier session could not be used to shape a later response to the child.

That provision is designed to prevent chatbots from maintaining an effectively unlimited memory of a child’s disclosures. It would not set the permissible period directly. The duration would be determined by the FTC through rulemaking after the bill became law.

The FTC would also be required to prohibit developers and deployers that know a user is a minor from configuring a chatbot with specified features intended to encourage prolonged or compulsive use.

Those features include rewards based on how frequently or how long a child uses the service, most notifications and push alerts, usage badges, unprompted chatbot messages, and indicators that mimic human interaction, such as typing bubbles or displays showing that the chatbot is online.

Developers and deployers could not configure a chatbot to advertise to a minor. They also could not generate a recommendation or endorsement whose credibility was materially affected by a financial relationship between the company and the seller of the recommended product or service.

Other provisions apply specifically to deployers, which the bill defines as entities that own, operate, or otherwise make a chatbot available to users. A deployer that knew a user was a minor could not process that child’s personal data to create a behavioral or personality profile. It also generally could not use the data to train a covered algorithm or transfer it to another party for that purpose.

The training restriction contains an exception allowing a deployer to process a minor’s information specifically to test for, identify, or address risks of harm to users.

After the FTC issued its rules, a deployer could otherwise process a child’s chatbot inputs only to generate a response using information from the current session within the permitted time or for those safety purposes.

The restrictions would apply when a company had actual knowledge, or knowledge implied by objective circumstances, that a user was a minor.

The bill expressly states that it would not require chatbot operators to collect age information they do not already collect or to introduce age-verification or age-gating systems.

The FTC and state attorneys general would enforce the measure. A parent or legal guardian could also sue a developer or deployer for violations of the bill’s design and privacy provisions and seek actual or punitive damages, legal costs, an injunction, or other relief a court considered appropriate.

The private right of action would belong specifically to parents and legal guardians, not to private plaintiffs generally.

COPPA 2.0

Markey’s agenda also includes the Children and Teens’ Online Privacy Protection Act, known as COPPA 2.0, a broader bipartisan proposal introduced with Republican Sen. Bill Cassidy of Louisiana.

COPPA 2.0 has passed both chambers of Congress in different forms. The Senate approved Markey and Cassidy’s standalone bill by unanimous consent on March 5, while the House included its version of COPPA 2.0 in the broader Kids Internet and Digital Safety Act, which passed 267-117 on June 29.

The legislation has not become law because the House and Senate have not yet approved the same final measure.

The bill would update the Children’s Online Privacy Protection Act of 1998, which Markey originally sponsored in the House. The current law generally focuses on children younger than 13. COPPA 2.0 would define teenagers as users who are at least 13 but younger than 17 and would extend important privacy protections to that group.

It would also prohibit operators from collecting personal information from users ages 13 through 16 without their consent, ban individual specific advertising to children and teenagers, and limit collection to information reasonably connected to the product, service or transaction requested by the young user.

The measure would also require companies to provide a mechanism for deleting information collected from a child or teenager and would revise COPPA’s knowledge standard.

Instead of applying only when an operator had actual knowledge that a user was a child, the law would also apply when the user’s age was fairly implied by objective circumstances.

Civil rights

Markey’s civil rights proposals address algorithms used to make or influence consequential decisions. His agenda argues that systems trained on historical records can reproduce existing patterns of discrimination, even when race, sex, or another protected characteristic is not expressly used as an input.

The Artificial Intelligence Civil Rights Act, reintroduced by Markey and Rep. Yvette Clarke, would regulate developers and deployers of covered algorithms used in commercial activities affecting employment, education, housing, credit, healthcare, and other significant opportunities.

The bill would prohibit a developer or deployer from offering, licensing, selling, or using a covered algorithm in a manner that causes or contributes to an unjustified disparate impact, otherwise discriminates or makes equal access to goods, services, or opportunities unavailable because of a protected characteristic.

Developers and deployers would have to conduct preliminary evaluations before using or offering covered algorithms for consequential actions. A developer would evaluate the potential harms associated with expected uses, while a deployer would examine whether its intended use might produce harm.

Independent audits would not be mandatory in every case. When a preliminary evaluation concluded that harm was plausible, the developer or deployer would have to engage an independent auditor for a more extensive pre-deployment review.

The review would examine the algorithm’s design, training, data, expected uses, limitations, and potential to cause harm or disparate impact.

After deployment, organizations using covered algorithms would have to conduct preliminary impact assessments annually. If an assessment identified harm resulting from the algorithm, the deployer would have to obtain a full assessment by an independent auditor.

Developers would separately review the assessment summaries they received from deployers each year.

The bill would direct the FTC to issue rules specifying when an individual must be offered the ability to opt out of an algorithmic process and have a consequential action undertaken by a human.

The commission would consider the seriousness of the decision, the risk of harm, the likely value of human involvement, and whether a human alternative would be practical.

The FTC would also determine when people must be allowed to appeal an algorithm-influenced decision to a human reviewer. The legislation would therefore not create an immediate, unconditional right to human handling or review for every decision involving an algorithm.

Nor would the bill immediately require companies to provide individualized explanations of every algorithmic result.

It would require public disclosures about developers’ and deployers’ practices, while directing the FTC to study the feasibility of requiring a mechanism through which affected people could learn whether and how a covered algorithm influenced them.

Algorithmic bias

Markey’s second civil rights measure, the Eliminating Bias in Algorithmic Systems Act, would focus on federal agencies rather than private companies. He reintroduced with Rep. Summer Lee.

The bill would cover agencies that use, fund, or procure a covered algorithm, finance or participate in its development, or oversee, regulate or advise on its development or use.

To qualify as covered, an algorithm would have to have the potential to materially affect a federally operated or funded program, an economic opportunity regulated by an agency, or rights protected by an agency.

Each covered agency would have to maintain a civil rights office employing experts and technologists focused on algorithmic bias, discrimination and other harms affecting protected groups.

Those offices would report to the congressional committees overseeing their agencies within one year of enactment and every two years afterward.

The reports would describe emerging risks, steps taken to mitigate harm, engagement with affected communities and recommendations for legislation or administrative action.

The Justice Department’s Civil Rights Division would also establish an interagency working group composed of the covered agencies’ civil rights offices.

With the exception of COPPA 2.0, none of the legislation has been advanced. Together, though, they define Markey’s approach to  limiting how companies use children’s information and mandatory scrutiny of algorithms that can affect civil rights and access to important opportunities.

Article Topics

age verification  |  algorithmic accountability  |  chatbots  |  legislation  |  U.S. AI policy  |  U.S. Government