By Benjamin Guggenheim

Not a subscriber? Sign up here to get this newsletter in your inbox.

• The Senate Commerce Committee has officially approved several children’s online safety and chatbot bills. But Democrats on the Commerce Committee ultimately stripped out preemption of state AI laws originally in the Chatbot Act.
• Efforts to pair children’s safety bills with a comprehensive AI package are all but dead. It’s unlikely we’ll see another shot at an AI package until 2027.
• I spoke with Paul Lekas of the Software & Information Industry Association about how companies are preparing for potential laws such as the Chatbot Act.

This is Benjamin Guggenheim. Welcome to WP Intelligence’s AI & Tech Brief, where we examine the transformative technology of artificial intelligence at the intersection of innovation, policy and power. Get in touch at Benjamin.Guggenheim@washpost.com or on Signal at ben_guggenheim.88.

The Lead Brief

The Senate Commerce Committee passed the landmark Kids Online Safety Act on Wednesday, along with several other bills establishing guardrails on chatbots.

But, in many ways, Congress is back to square one on negotiating a comprehensive AI package.

First, what happened: The Senate Commerce Committee approved legislation requiring chatbots to delete data on interactions with minors and disclose to users that they are interacting with AI.

The approved Children’s Artificial Intelligence Toy Safety Act of 2026 likewise directs the National Academies of Sciences, Engineering, and Medicine to study the emotional and mental impacts of children interactions with “AI toys” — and directs the FTC to submit a plan to Congress within two years with legislative recommendations based on the findings.

The approved Chatbot Act would establish family chatbot accounts for children under 13, and would require parental consent for teens to use chatbots. The bill bans targeted advertising to minors.

The path forward: Given that the House has already passed its version of a children’s safety package, called the Kids Act, the expectation is that the House and Senate would go to “conference” to reconcile the bills — pending the upper chamber taking them up and likely passing the Senate Commerce bills.

But the House Republican majority and Democrats are still opposed to the “duty of care” language in the Senate KOSA bill, which imposes an obligation on tech companies to mitigate foreseeable harm to users.

“Whether there is a negotiation to be had will depend on whether the Senate is willing to drop the duty of care or not,” as one senior House GOP aide, who spoke on the condition of anonymity, bluntly put it to me.

The preemption fight: There was an overlooked development of the Senate Commerce markup: Language in the Chatbot Act that would have preempted similar state AI laws governing chatbots was ultimately stripped out at the request of Democrats.

The language now reads: “Nothing in the Act … shall be construed to prohibit … any State law, requirement, or regulation that is at least as protective of users of AI chatbots as this Act.”

Republicans’ and billionaire David Sacks’ effort to achieve a level of preemption — even after consideration of most of the AI bills was taken off the table until September — has come up short again.

FILE - The OpenAI logo is seen on a mobile phone in front of a computer screen. (AP Photo/Michael Dwyer, File)

Stepping back: After passage of these bills, the effort to pair children’s online safety legislation with an AI package appears to be officially dead. The political rationale for connecting the two issues was to get Sen. Marsha Blackburn (R-Tennessee) and a swath of conservative family groups on board with an AI package.

But now, barring something surprising, it looks almost certain that lawmakers will need to formulate an AI package on a bipartisan basis in the upcoming 120^th Congress — with strong concessions to Democrats on policies addressing job displacement, data centers and AI model testing.

If the Chatbot Act is enacted, businesses developing and deploying chatbots in education, medicine, tech and a myriad other areas will have to contend not only with federal laws but also state laws in California, New York, Oregon and Nebraska.

To understand the landscape, I spoke with Paul Lekas, executive vice president for global public policy and government affairs at the Software & Information Industry Association (SIIA), which represents nearly 400 companies operating in the information industry.

This interview has been edited for clarity and brevity.

Q: You have written the committee on behalf of SIIA, urging lawmakers to refine the definition of covered interfaces to companion chatbots. What other interfaces are you concerned could be swept in?

A: As a preliminary matter, the issue of youth and chatbots is really important, and we care about it.

There are a number of states that have advanced chatbot bills already. There are differences among the different state models. But one of the things that we’ve seen in the states that I think is valuable is that the legislation that focuses on companion chatbots tries to define that space more clearly. The Chatbot Act starts with a very broad definition of a chatbot — it is something that accepts user input and engages in interactive conversations with a user — and then it has a series of carveouts.

The carveout for some of the typical enterprise uses like customer service, shopping assistance and business operational purposes is conditioned with the phrase “limited to contextual replies” or “to a narrow, specified purpose.”

We feel like that is going to be too limited. A customer service chatbot may do more than the ones that we’ve been used to over the past several years.

We’re seeing that companies across the board are looking at ways to incorporate AI into all kinds of different interactions they have, and so that’s definitely a concern.

Q: Your association includes several companies that work in the ed tech space, and you write that the exception for educational chatbots that are “closed, curriculum bound” (as defined in the Chatbot Act bill) is too narrow. Can you explain that?

A: We speak in the letter to those at a very high level — so, tools that help to build skills and provide guidance on vocational and craft training.

I think the way it’s defined in here would capture certain types of closed-system chatbot tools that might help you with learning mathematics, potentially even in a personalized way — but, as you cross disciplines and go beyond the specific confines of a curriculum, you still remain within the educational domain.

We work with a number of companies that do this. When we talk about ed tech, we’re very specific about tools that are purpose-built [and] that comply with [the Family Educational Rights and Privacy Act] and other requirements [concerning] student privacy that are used in the classroom — pursuant to a contract with a district or a state educational authority or a school.

Q: What does it mean for the companies that you represent that the preemption language has basically been stripped out of the chatbot bill?

A: I think the preemption debate has taken on a life of its own, and it’s a proxy for a number of other debates that people have about what legislation should actually look like — and how we should actually be protecting youth.

Our general view is preemption is the wrong way to think about it: We are looking for federal law. We are looking for consistency across the country. And we want to protect youth safety across the country.

There are those who are concerned that a federal law will be too weak and that we need more stringent state laws. That presents a number of challenges for youths who may [experience] different experiences as they leave the state and go to college.

On the company side, it is an implementation [and] engineering challenge to provide different experiences to everybody. It is one thing to implement an age gate in a particular geographic area — which, as we know, can be circumvented through the use of VPN. Age-gating is different than providing a fundamentally different experience everywhere, and that is a really challenging engineering problem.

Q: There are limitations put on data retention in the federal chatbot legislation, but you point out that companies may need to retain data longer to develop classifiers to detect interactions with minors. Can you explain the considerations when policymakers think about data retention?

A: There’s universal agreement that child sexual abuse imagery material is fundamentally bad, and we do not want it on the internet. We want to limit the ability for AI tools or the internet to generate that [type of content] to disseminate and so forth.

However, in order to be able to identify that kind of material on the back end, [there is a need] to be able to evaluate what those markers are so that we can detect that, and we can prevent anybody from seeing that kind of material.

Likewise, it is valuable to have limited access to data in order to make sure that the models are being trained in the right way to prevent the kinds of situations that a bill like the Chatbot Act wants to prevent.

It should not be used to train a general model, but it should be available for the sole purpose of training classifiers that are completely depersonalized.

This newsletter is published by WP Intelligence, The Washington Post’s subscription service for professionals that provides business, policy and thought leaders with actionable insights. WP Intelligence operates independently fromThe Washington Post newsroom. Learn more about WP Intelligence.