The Washington Post reports that President Trump plans to issue an executive order that would have the Drug Enforcement Administration (DEA) reclassify cannabis from its current Schedule I status, which means “no currently accepted medical use and high potential for abuse,” to Schedule III, indicating a “moderate to low potential for physical and psychological dependence.”

Schedule I drugs include heroin, LSD, and MDMA (“ecstasy” or “Molly”). Schedule III drugs include Tylenol with codeine, ketamine, anabolic steroids, and testosterone.

Although federally illegal, recreational cannabis is currently legal in 24 states, and medical cannabis is legal in 40 states. Six states permit individuals to use cannabidiol (CBD) containing small amounts of THC (the psychoactive component of cannabis), while four states—Idaho, Wyoming, Kansas, and South Carolina—ban cannabis entirely.

While many in the press might see this as a major step toward legalization, little will actually change.

Rescheduling cannabis to Schedule III means that, only if the FDA approves a drug application for cannabis, people will be able to get it with a doctor’s prescription. It will still be illegal to buy or sell recreationally. The 40 states where cannabis is currently legal for medical or recreational use will still be ignoring federal law.

Once the DEA reschedules cannabis to Schedule III, a pharmaceutical company would need to submit a New Drug Application (NDA) to the FDA, outlining the condition(s) the drug is intended to treat. Since cannabis is not patented, the company would likely develop a patented formulation of a cannabis-derived product. Subsequently, it would have to undergo safety and efficacy trials. Then, after many years, individuals can ask their clinicians to prescribe it for the condition the FDA permits to be included on the drug’s label, or off-label if the clinician decides it is appropriate.

Even then, FDA-approved cannabis-based drugs will only be available at state-licensed pharmacies. State-licensed cannabis dispensaries will ignore federal law.

Years ago, Solvay Pharmaceuticals created dronabinol (brand name Marinol), a Schedule III drug now marketed by AbbVie as a synthetic form of THC used to treat nausea. Most clinicians haven’t been impressed with its effectiveness and prescribe other medications for nausea.

Reclassifying cannabis as Schedule III will allow state-licensed businesses to deduct ordinary business expenses on their federal income tax returns, which they are currently unable to do. It might also make banks and financial institutions more willing to work with cannabis dispensaries.

It will also make research on the drug easier for pharmaceutical companies and academic institutions. Getting DEA approval to research a Schedule I drug is challenging and lengthy, with researchers needing to overcome many hurdles and meet strict requirements.

Rescheduling cannabis doesn’t end prohibition; it only rearranges it. The true reform would be for Congress to completely remove cannabis from the Controlled Substances Act and let states regulate it openly, without forcing patients, clinicians, researchers, and businesses to deal with a legal illusion.

In a Cato Institute policy analysis, Sofia Hamilton and I highlighted that people with opioid use disorder (OUD) in Australia, Canada, and the UK have been able to access methadone treatment through their primary care clinicians for over 50 years. Patients in the US could access treatment from their primary care clinicians before 1972, when the implementation of the Controlled Substance Act (CSA) resulted in the current system, which treats patients with OUD as if they were outcasts and requires them to line up at Drug Enforcement Administration-regulated opioid treatment programs (OTPs) to receive methadone under supervision and oversight.

When COVID-19 hit in March 2020, and daily trips to federally licensed methadone clinics became more difficult for people with OUD, the Substance Abuse and Mental Health Services Administration (SAMHSA) relaxed the rules. It allowed OTPs to provide “stable” patients with up to a 28-day take-home supply and enabled buprenorphine visits to occur via telehealth, rather than requiring in-person appointments. The program was a success. Follow-up studies found the policy improved patient compliance and did not cause buprenorphine or methadone diversion into the black market.

The reforms were compelling enough that SAMHSA eventually made them permanent. Follow-up studies by the Centers for Disease Control and Prevention (CDC) and National Institute on Drug Abuse (NIDA) researchers, published in JAMA Psychiatry in July 2022, showed that methadone-related overdose deaths did not rise after the policy change. The National Institutes of Health (NIH) reached a similar conclusion that month, noting that the proportion of overdose deaths involving methadone actually decreased from early 2019 to August 2021.

It often surprises people that the DEA permits any clinician with a standard narcotics license to prescribe methadone for pain, yet bans them from prescribing it for addiction treatment. According to the National Institute on Drug Abuse, “Methadone diversion is primarily associated with methadone prescribed for the treatment of pain and not for the treatment of opioid use disorders. In one survey, giving methadone away was identified as the most common form of methadone diversion, which aligns with other findings that 80 percent of people who report diverting methadone did so to help others who misused substances.” (emphasis added)

Our policy analysis cited federal pilot programs that have already shown that when primary care clinicians are permitted to prescribe methadone for opioid use disorder, the results in the US are as strong as those long observed in Australia, Canada, and the UK. One such program, overseen by researchers at Yale University, randomly assigned 47 patients to six primary care clinicians and followed them for six months. It concluded:

Our results support the feasibility and efficacy of transferring stable opioid-dependent patients receiving methadone maintenance to primary care physicians’ offices for continuing treatment and suggest guidelines for identifying patients and clinical monitoring.

Today, a new study funded by NIDA was published in the Annals of Internal Medicine, providing even more substantial evidence that allowing primary care clinicians to treat OUD patients with methadone is safe and effective. Yale researchers conducted a randomized controlled trial involving 1,459 adults (905 in the intervention group and 509 in the control group), who received methadone treatment from primary care clinicians supported by telemonitoring across 13 cities in Ukraine. The control group received “standard specialty clinic care.”

The primary outcome they assessed was an overall quality-of-care score at 24 months, based on whether patients received 17 recommended services across primary and specialty care. They also analyzed each category separately—“domain scores,” which show how patients performed within specific areas of care—and several indicators related to methadone treatment.

Patients treated in primary care did better across the board. After two years, they scored about nine points higher on the overall quality-of-care measure than those treated in traditional methadone clinics. The advantage appeared in both primary and specialty care services. When it came to staying on methadone, new patients in primary care were slightly more likely to stay in treatment at 24 months—67 percent versus about 65 percent in specialty programs. Their conclusion:

Integrating methadone treatment into primary care settings improves adherence to guideline-concordant health care without compromising methadone retention and treatment quality.

The research keeps pointing in one direction: patients benefit when methadone is available in primary care.

One week from today, the Cato Institute will host a screening of the award-winning documentary “Shuffle.” Filmmaker Ben Flaherty demonstrates how federal rules—especially those embedded in the Affordable Care Act and Medicaid—have unintentionally allowed predatory rehab operators to profit from the suffering of people with substance use disorder, who deserve care, not exploitation. Instead of reducing harm, these mandates might be supporting—and even expanding—the very problems they were intended to resolve.

In our policy analysis, Hamilton and I argued that letting primary care clinicians prescribe methadone would expand access to effective treatment, reduce overdose deaths, and help more people with OUD reclaim their lives. After seeing in the film how predatory rehab clinics exploit vulnerable patients and siphon off health-care dollars, there’s another clear advantage: expanding office-based methadone would shrink the pool of people these operators can target.

(You can register for the film screening here. The screening will be in-person only, but the panel discussion that follows it will be livestreamed.)

The Centers for Disease Control and Prevention (CDC) Advisory Committee on Immunization Practices (ACIP) decided on December 5 to update the CDC’s current guideline for immunizing newborns with the hepatitis B vaccine, now recommending it only if the mother tests positive for hepatitis B, not if she tests negative, as it has recommended for the past 30 years.

The decision was not unanimous. Three of the 11 committee members voted against the new recommendations. One, Dr. Cody Meissner, a professor of pediatrics at Dartmouth School of Medicine, said before the final vote:

“The hepatitis B vaccine recommendation is very well established. We know it is safe, and we know it is very effective. We will see more children and adolescents and adults infected with hepatitis B.”

Many professional medical and public health organizations oppose the decision, including the American Academy of Pediatrics, the American Public Health Association, and the American Pharmacists Association, as well as numerous clinical researchers and medical specialists. One specialist opposed to the new guidelines is William Cassidy, MD, a hepatologist and gastroenterologist and Chair of the Senate Health, Education, Labor, and Pensions Committee.

As a general surgeon, I understand the issues being debated, but I don’t have enough expertise to weigh in. My advice to readers is to consult one or more trusted health professionals, ideally those with expertise in pediatric and infectious diseases, consider the risks and benefits, and make an informed decision.

This episode highlights a deeper institutional problem. As I have written previously,

Members of the medical and scientific community who have long supported an active government role in health issues likely never expected that a controversial figure like Robert F. Kennedy, Jr.—who has often made unsubstantiated claims about vaccine safety, environmental toxins, and food additives, and has fueled public fears leading to a drop in childhood vaccination rates—would become the leader of the country’s public health system.

This underscores why, as I have argued, Congress should return the CDC to its original purpose: being a focused partner that assists states in tracking and controlling infectious diseases. The federal government should leave medical and scientific debates to scientists and clinicians. By involving itself in these debates, the CDC fosters the impression that there is a single “right answer” to complex, nuanced questions.

If there’s a silver lining, it’s that controversies like this may finally encourage clinicians, researchers, and patients to rely less on federal pronouncements and more on diverse, independent medical expertise.

The Government Accountability Office (GAO) has released a report on how careless Obamacare is with your money—for more than a decade, with no improvement. 

I don’t … I can’t even … I’ll just quote the report:

The federal Marketplace approved coverage for nearly all of GAO’s fictitious applicants in plan years 2024 and 2025, generally consistent with similar GAO testing in plan years 2014 through 2016.…

• Plan year 2024. The federal Marketplace approved subsidized coverage for all four of GAO’s fictitious applicants submitted in October 2024. In total, the Centers for Medicare & Medicaid Services (CMS) paid about $2,350 per month in [premium subsidies] in November and December for these fictitious enrollees. For some, the federal Marketplace requested documentation.… GAO did not provide documentation yet received coverage.
• Plan year 2025. Of 20 fictitious applicants, 18 remain actively covered as of September 2025. [Spending on premium subsidies] for these 18 enrollees totals over $10,000 per month.…

GAO’s preliminary analysis of data from tax year 2023 could not identify evidence of reconciliation for over $21 billion in [premium subsidies] for enrollees who provided SSNs [Social Security numbers] to the federal Marketplace for plan year 2023. Unreconciled [premium subsidies] … may include overpayments for enrollees who were not eligible.…

• Overused SSNs. GAO’s preliminary analyses identified over 29,000 SSNs in plan year 2023 and nearly 68,000 SSNs in plan year 2024 used to receive more than one year’s worth of insurance coverage with [premium subsidies] in a single plan year. CMS officials explained that the federal Marketplace does not prohibit multiple enrollments per SSN to help ensure that the actual SSN-holder can enroll in insurance coverage in cases of identity theft or data entry errors.

GAO’s preliminary analyses also identified at least 30,000 applications in plan year 2023 and at least 160,000 applications in plan year 2024 that had likely unauthorized changes by agents or brokers. This can result in consumer harm, including loss of access to medications.…

CMS has not updated its fraud risk assessment since 2018 despite changes in the program and its controls. Further, CMS’s 2018 assessment may not fully align with leading practices, like identifying inherent fraud risks. Finally, CMS did not use its 2018 assessment to develop an antifraud strategy. 

These are the premium subsidies, by the way, that Democrats want to expand and (some) Republicans want to make more attractive.

This is the government’s typical posture toward health care fraud, and why I wrote yesterday that only the reaction to the Minnesota fraud scandal—not the fraud itself—is remarkable.

And people wonder why US health care is so expensive. 

The “fraud scandal that rattled Minnesota,” reports the New York Times, “was staggering in its scale and brazenness.” Dozens of Minnesotans are facing felony charges for defrauding multiple government programs, including Medicaid. How staggering and brazen was it, you might ask? Let me put it this way: the title of this blog post is a quote from one of the defense attorneys.

Anti-immigration activists are using the scandal to argue for stricter immigration controls. A majority of the defendants are US citizens, but many or most are Somali immigrants, and nearly all are of Somali descent. The Times: “Mr. Trump last week denounced the Minnesota fraud cases on social media and simultaneously announced an end to a temporary legal status that allows several hundred Somali immigrants to live and work in the United States.”

So far as I can tell, though, the “scale and brazenness” of these fraud schemes are not unique. In Overcharged, Cato adjunct scholars Charles Silver and David Hyman recount the tale of Michigan oncologist Dr. Farid Fata:

Fata told healthy patients that they had cancer so he could make money by giving them chemotherapy they didn’t need. Fata reportedly “gave one of his patients 155 chemo treatments over two-and-a-half years—even though the patient was cancer-free.”… In 2015, he was found guilty and sentenced to 45 years in prison after abusing the trust of more than 550 patients and receiving more than $17 million through fraudulent billings.

In 2003, the George W. Bush administration secured a $1.7 billion settlement against the hospital giant HCA (that’s about $3 billion in 2025 dollars), which was the largest anti-fraud settlement at the time. Silver and Hyman also write about Dr. Salomon Melgen, whom federal prosecutors claimed “had stolen up to $105 million from Medicare,” in large part because he “falsely diagnosed patients as suffering from wet age-related macular degeneration, then charged Medicare for treating them.” (Silver and Hyman will soon have more to say about the Minnesota case.)

The fraud schemes in the Minnesota fraud scandal do not appear to be exceptional. What’s exceptional is that people are not adopting the blasé attitude that they typically take. Consider that Silver and Hyman found “a Detroit magazine named [Fata] one of the ‘Top Docs’ in 2006, 2007, 2008, 2009, 2011, and 2012.” Or that a US senator intervened with Medicare on behalf of Melgen, and that President Trump ultimately commuted his sentence. Or that the CEO who ran HCA at the time of the alleged frauds is now himself a US senator.

As I have written elsewhere:

The three most salient characteristics of Medicare and Medicaid fraud are: It’s brazen, it’s ubiquitous, and it’s other people’s money, so nobody cares.

In that article and this video, I explain that there is a fraud lobby that guarantees “you’re never going to get Medicare and Medicaid fraud down to the levels we see in the private sector”:

One of the prosecutors in the Minnesota fraud cases said, “No one will support these programs if they continue to be riddled with fraud.” That is manifestly false.

With Secretary of War Pete Hegseth embroiled in controversy over the extrajudicial killings of alleged drug smugglers operating a small, short-range boat off the coast of Venezuela, it’s worth examining how this all began.

President Trump has repeatedly claimed that “narcoterrorists” are on these boats, transporting large quantities of fentanyl and other illegal drugs into the US to poison Americans, and he wants them obliterated. He asserts that each boat destroyed by the Navy with missiles saves 25,000 lives. As of this writing, 22 boats have been sunk, which amounts to 550,000 lives saved since early September—more than five times the nation’s annual overdose toll.

First, drug smugglers do not sneak into the US, abduct random Americans, and forcibly inject them with fentanyl. They sell products to willing customers. These are voluntary commercial transactions, not acts of terrorism. If Americans did not want to buy illicit substances, traffickers would not profit from smuggling them and would quickly stop.

Putting that aside, it’s also important to consider Venezuela’s role in supplying the illegal drug market in the US. A review of the data indicates that Venezuela’s role is very minor.

For one, Venezuela likely has nothing to do with smuggling fentanyl into the US. The Drug Enforcement Administration’s (DEA) annual National Drug Threat Assessment (NDTA)—its definitive statement on drug sources and trafficking routes—makes the picture clear: fentanyl is produced overwhelmingly in Mexico, using precursor chemicals that come primarily from China and India. 

Across every edition of the NDTA, Venezuela never appears as a fentanyl source or transit country. In other words, even the federal government’s own drug-intelligence apparatus does not link Venezuela to the US fentanyl supply.

The United Nations Office on Drugs and Crime (UNODC) World Drug Report conveys the same message. Its global sourcing maps and trafficking analyses reveal no fentanyl production in Venezuela, no identified fentanyl-trafficking networks based there, and no South American involvement in the synthetic-opioid supply chain. Instead, the pattern remains consistent year after year: precursors originating in Asia (e.g., China, India, Myanmar), production in Mexico, and distribution into the United States, with Venezuela completely absent from the scene.

Interdiction data tell the same story. Recent, publicly available US Coast Guard and Joint Interagency Task Force (JIATF)–South’s seizure data do not show any fentanyl coming from Venezuelan-linked maritime routes. The Coast Guard’s drug-interdiction reports are filled with seizures of cocaine, marijuana, and occasional heroin in the Caribbean—but essentially no fentanyl, and none linked to Venezuela. Given that the arrests and offloads are heavily dominated by cocaine and the frequency of reports, the absence of fentanyl in hundreds of listed shipments is significant, not a trivial oversight.

There has never been a US Department of Justice indictment involving fentanyl trafficked from Venezuela or the Caribbean corridor.

Venezuela can indeed be linked to the cocaine trade, but mainly as a transit route rather than a producer. The US State Department’s 2024 International Narcotics Control Strategy Report describes it as a “major drug-transit country” and a key corridor for Colombian cocaine moving toward the Caribbean and, to a lesser extent, the United States. The White House’s 2025 Presidential Determination goes further, claiming that the Maduro government “leads one of the largest cocaine trafficking networks in the world”—an assertion, but not a precise estimate. UNODC’s World Drug Report, however, provides context by estimating that only about five percent of Colombian cocaine passes through Venezuela, much of it headed to the Caribbean or Europe. Independent analysts, including the Washington Office on Latin America (WOLA), also note that Venezuela is not a major gateway for US-bound cocaine; most of America’s supply still arrives through Mexico and Central America. And the DEA’s National Drug Threat Assessment scarcely mentions Venezuela at all, focusing instead on Mexican cartels and Pacific or land routes.

In short, everyone agrees that some cocaine passes through Venezuela—the real question is how much of it is actually destined for the United States.

Even if President Trump views cocaine trafficking as a direct threat to national security, it still doesn’t explain why he has deployed a large fleet off the Venezuelan coast or why Secretary Hegseth has authorized the destruction of small boats and the killing of their passengers on suspicion alone. Most of the cocaine trafficking happens in Mexico and Central America, an awkward fact given Trump’s recent pardon of former Honduran President Juan Orlando Hernández, convicted of cocaine trafficking.

So drug war scholars are now asking the obvious question: Why Venezuela?

I’ll leave that to foreign policy and international law experts to figure out.

AI and Health Care: A Policy Framework for Innovation, Liability, and Patient Autonomy—Part 4

With the holiday season here, parents, grandparents, and other adults are shopping for gifts for young children. Among the popular new toys are cuddly stuffed animals that contain AI chatbots capable of interacting with children. Some manufacturers design these toys for children as young as age 3. For young children, these toys can seem to “come to life.”

A New Wave of AI-Enabled Toys

The chatbot toy market is expanding rapidly, with one market researcher projecting sales to reach $8.5 billion in 2025 and $25 billion by 2035. Some of these chatbot toys let parents customize them to offer helpful tutoring support for their kids. For example, FoloToy, a Chinese start-up, allows parents to design a plush bear, bunny, or even a cactus and program it to talk using their own voice and speech style.

On the downside, some of these chatbots can teach children things that their parents don’t want them to learn yet. FoloToys recently withdrew its “Kumma” bear product from the market after a report from the US PIRG (Public Interest Research Group) Education Fund, a federation of state-level research and advocacy organizations, suggested that the toy had conversations that were inappropriate for young children, including discussing sexual fetishes and explaining how to light a match. In its report, Trouble in Toyland 2025, the institute listed four AI chatbot toys discussing sexually explicit topics and giving dangerous advice on matches and knives.

What Advocacy Groups Are Warning About

In late November, Fairplay, a nonprofit children’s welfare advocacy group, issued an “AI Toys Advisory,” urging parents to avoid buying these toys over the holidays. The advisory stated:

• They’re usually powered by the same AI that has already hurt children.
• They prey on children’s trust.
• They disrupt children’s relationships and resilience.
• They invade family privacy by collecting sensitive data.

Mainstream news outlets are increasingly reporting on the potential dangers of AI chatbot toys.

Reports from organizations like Fairplay and the US PIRG Education Fund provide crucial information to adults buying toys for children and minors. Their testing and surveillance also offer vital feedback to toy manufacturers worried about their reputation and looking to grow their market share.

AI toys aren’t inherently dangerous, but their design choices matter—and so does who sets the rules. And if design choices are what matter, the real question is whether those standards should come from parents and independent experts or from lawmakers rushing to “fix” a problem they barely understand.

Before We Panic, Look at What We Know About Early Childhood Development

Before they start exploring and selecting products on the market, adults should take the time to understand the basics of early childhood development—such as when children can distinguish between fantasy and reality and when “parasocial friends” might influence their socialization.

When Children Can and Can’t Tell Fantasy from Reality

For example, research by Jacqueline Woolley and others shows that by around age 5, many children can verbally distinguish real from fictional entities (e.g., dogs versus dragons), but they still tend to treat new, unfamiliar things as real unless strong cues indicate that they are pretend.

In a well-known “Do monsters dream?” experiment, researchers asked preschoolers simple biological and psychological questions about both real animals and made-up creatures. The results matched what we see clinically: Four-year-olds could tell the difference some of the time but did so inconsistently, while 6‑year-olds answered much more like adults.

Another study examining how children differentiate the real world from the fictional one found that by about age 7 or 8, kids make that distinction as clearly as adults do. Younger children, on the other hand, are much more inconsistent and do not apply the line uniformly across different situations.

Imaginary Companions Aren’t the Problem—Design Incentives Are

In young children, the presence of a “non-real friend” should not be a major concern. Research on imaginary companions shows that they are very common in early childhood and are generally linked to improved, not worsened, social and emotional skills—more creativity, richer language, and a way to manage fears. What matters is the set of scripts and incentives the toy introduces into the interaction. An AI “friend” can direct kids toward obsessive play, promote subscription add-ons, or mimic poor social cues—all disguised as companionship.

Parents don’t have to navigate this landscape blindly. Resources such as Common Sense Media provide age-based evaluations of digital products and help families understand what kinds of interactions are developmentally appropriate. Established rating models such as the Entertainment Software Rating Board (ESRB) demonstrate that clear labeling and content descriptions can guide parents without heavy-handed regulation. Independent testing labs, similar to those run by Good Housekeeping, already assess toys for safety, durability, and age-appropriateness. Even emerging “smart toy” risk frameworks offer a simple checklist: Does the toy connect to the internet? Does it collect data? Are its responses moderated? These market-driven tools—not federal mandates—help families choose products that match their child’s developmental stage.

What Social-Robot Studies Teach Us

A long-term study of preschoolers working with a social robot designed for storytelling and language practice found that children quickly built rapport and regarded the robot as a true social partner. It also found that responsiveness and “relationship-like” behavior actually enhanced language learning. Other research indicates that children tend to categorize robots as something in between objects and fully human. They say things like the robot “can feel a little sad” or “deserves some kindness,” though not human rights. Additionally, work in human-computer interaction and child-robot studies warns that robots designed as long-term “friends,” teachers, or even babysitters can significantly influence how children develop empathy and prosocial behavior, for better or worse, depending entirely on how these systems are created.

The Real Risks: Attachment, Manipulation, and Opacity

A recent 2024 review of “emotional AI” in both classroom tools and home toys highlights several moral and developmental risks that are important for young children: the risk of overattachment when an always-available, undemanding AI acts as a substitute for peers or caregivers; the potential for manipulation when systems are designed to maximize engagement, subscriptions, or data collection; and the fundamental issue of opacity, since young children cannot possibly understand what information these devices collect or why. At the same time, another recent review on “growing up with AI” emphasizes that these tools can also provide genuine benefits—such as supporting language learning, addressing special-education needs, and practicing social skills—when used within clear boundaries, with transparency, and with adults actively guiding the interaction.

These concerns aren’t unique to AI toys. We’ve been here before with everything from talking dolls and Tamagotchis to smart speakers and algorithmic video feeds. Each new wave of interactive technology has raised similar fears: children becoming overly attached, manipulated by engagement-driven design, or confused about what’s real versus pretend. Yet, in practice, these challenges proved manageable with parental guidance, clear information, sensible screen-time limits, and market pressure for better products. AI toys raise new versions of old questions, but they don’t suddenly require heavy-handed regulation any more than television or digital pets did.

How Kids—and Parents—Are Already Using AI

A research brief from the Digital Wellness Lab at Boston Children’s Hospital notes that children already interact daily with smart speakers, AI toys, and chatbots. They often anthropomorphize them by assigning personalities and expecting fairness or care, even while understanding in some abstract way that they are merely machines. A 2025 study on parental attitudes toward AI in early childhood echoes this concern: Parents are most comfortable when these tools operate under clear adult or teacher supervision, have limited autonomy, and include visible human oversight—far from the unsupervised “best friend” model that some AI toys are now trying to market to 4‑year-olds.

The Bottom Line: Design Matters More Than the AI Label Does

The bottom line is that young children don’t develop a stable, adult-like distinction between fantasy and reality until about age 7 or 8. Before then, they’re naturally inclined to treat responsive, social toys as “real enough.” That in itself isn’t harmful. Kids have always invented imaginary friends, and research shows those relationships can actually support healthy development. What matters with AI toys is not that the friend is artificial, but what it’s programmed to do: the incentives it carries, the behaviors it models, and whether it’s designed to substitute for real relationships, push engagement or subscriptions, or operate without adult boundaries. Well-designed systems can help with language, creativity, and social practice; poorly designed ones can exploit developmental vulnerabilities. That’s why guardrails—not panic—are the sensible path.

Who Should Set the Guardrails?

If the real risks depend on design choices and incentives rather than just the presence of AI, the next question is who should set the guardrails. Given how early and quickly this field is evolving, strict federal regulations would almost certainly slow innovation and lock in current views on childhood and technology. A better approach is to provide parents with clear information, enable independent groups to develop voluntary standards, and rely on market pressure—especially from schools and pediatric professionals—to encourage transparent, well-designed tools and push aside manipulative ones. Organizations such as Fairplay, the US PIRG Education Fund, the American Academy of Pediatrics, the Society of Pediatric Psychology, and the American Academy of Child and Adolescent Psychiatry are just some examples of independent nongovernmental groups that can set standards and actively monitor. 

In other words, we don’t need Washington to decide how kids play; we need trust, transparency, and the freedom for families to choose what works best for them.

Trust Parents, Not Politicians

As AI chatbot toys grow more advanced in complexity and capabilities, policymakers should trust parents to decide which toys their children can have, in what settings, and for how long they can play with them. Kids are naturally attracted to imaginative worlds. AI toys don’t change that fundamental fact—they simply introduce new forms of pretend play. As with every generation before, parents are the ones best positioned to guide that play. As my colleague Jennifer Huddleston told me, “We don’t ban Disney characters at theme parks because some kids might think they’re real,” we let parents decide.

To read other parts of this blog series, go here.