AI Is Becoming America’s Sex Educator. Is It Ready?
Anthropic’s Daniela Amodei invited parents to tell AI companies how they can do better. As a mother and sexologist, I have an answer.
On a flight home from Germany, somewhere between the meal service and my vain attempt to get comfortable in coach, I watched the 2026 documentary The AI Doc: Or How I Became an Apocaloptimist.
Near the end, Anthropic cofounder and president Daniela Amodei essentially invited parents to ask AI companies—including Anthropic—how they could do better.
Okay, Daniela. I have something to say.
(And although Daniela’s invitation prompted this essay, my answer is not only for Anthropic. It is for every company building AI systems people turn to for information, reassurance, advice, and connection.)
Let me begin with the boundary that should not be controversial…
AI must never sexualize children, facilitate abuse or grooming, create exploitative material, or position itself as a child’s romantic or sexual partner. Period.
Protecting young people requires more than blocking anything related to sexuality. Sometimes the safer response is an accurate, age-appropriate answer to a question about a body, a relationship, or an experience that did not feel right.
I come to this conversation as a mother. I am also a sexologist, sexuality educator, former adjunct faculty member, former HR Director who wrote policy and investigated sexual harassment and other misconduct, author, and, most recently, an AI product builder.
For more than fifteen years, I have worked with the questions people find hardest to ask out loud: bodies, consent, coercion, sexual development, pleasure, shame, relationships, and that quiet internal alarm that says, Something about this doesn’t feel right.
None of those questions are going away, but some of the trustworthy places young people could take them are.
When trustworthy information disappears, the questions remain
In June 2026, the Department of Health and Human Services issued new funding requirements for the federal Teen Pregnancy Prevention Program. A few days later, it canceled the vast majority of active grants midway through projects that had been funded to continue through 2028.
A coalition including SIECUS, Planned Parenthood of the Heartland, Hennepin County, and King County has now sued HHS. According to the complaint and supporting materials, affected organizations have already suspended programs, interrupted research, disrupted partnerships with schools and healthcare providers, and begun dismantling work that was serving young people.
I learned about the lawsuit through Anne Hodder Shipp, a sex educator and board member of SIECUS, one of the organizations bringing the suit. Their post sent me straight to the underlying documents.
The lawsuit argues that the new requirements replace a congressionally mandated evidence-based framework with abstinence-before-marriage programming, medically inaccurate claims about contraception, and restrictions that exclude or erase some of the young people most likely to need support.
Those allegations are now being litigated, and the administration may offer its own explanation for the changes.
But the interruption of active programs is not theoretical.
And these programs were not simply handing out pamphlets telling teenagers not to become pregnant.
For more than fifteen years, the Teen Pregnancy Prevention Program has developed, tested, evaluated, and expanded approaches to adolescent sexual health. It helped create an evidence base for working with young people in foster care, the juvenile justice system, rural communities, LGBTQ+ communities, and other populations often missed by one-size-fits-all programs.
A 2023 article in Prevention Science found that more than half of the programs then included in the federal evidence review had been developed or tested through the TPP Program. Its innovation networks had also directly involved tens of thousands of young people, parents, caregivers, and community members in designing and evaluating better approaches.
This was not a static curriculum.
It was a national system for iterating and learning what actually helps. The article abstract is available here.
Whatever your politics, the systems problem should concern you.
Nobody had to sit in a room and say, “Let’s drive teenagers toward chatbots.” It happens anyway.
Reduce access to evidence-based education. Make certain subjects harder for teachers and healthcare professionals to discuss. Leave parents with the same incomplete education many of us received. Add shame, fear, family conflict, religious pressure, concerns about confidentiality, or the possibility that the trusted adult is also the source of danger.
Then place an always-available conversational system in nearly every young person’s pocket.
(You can see where this is going.)
AI is easy to ask
Young people do not stop wondering about their bodies because an adult would prefer they wait.
They do not stop developing crushes, exploring identity, feeling sexual sensations, finding porn, experiencing pressure, or trying to figure out whether something that happened to them was okay.
They take those questions wherever it feels easiest and safest to ask.
This used to be mainly Google. Increasingly, it means AI. It is always available, seemingly private, and unlikely to react with embarrassment or judgment.
“Seemingly private” deserves emphasis.
A young person may experience a chatbot conversation as confidential even when the platform collects, stores, reviews, or uses that information in ways the user does not understand. Real privacy protections and clear explanations of their limits must be part of any meaningful child-safety design.
Still, the appeal is obvious.
You don’t have to make eye contact with AI.
You don’t have to wonder whether it will look shocked.
You don’t have to know the correct anatomical term before you begin.
You can type the question you are too embarrassed to say, delete it from your screen, and pretend you never asked.
And this is already bigger than sex education.
AI systems are becoming confidants, relationship advisers, emotional supports, friends, and sometimes romantic or sexual companions.
In a 2025 national survey, Common Sense Media found that nearly three-quarters of American teens had used AI companions. A substantial minority had used them for social interaction or relationships. One-third had chosen an AI companion over a human for a serious conversation. A quarter had shared personal information with one. The findings call for precaution, not panic, but they make clear that this is already happening.
To be clear, this did not begin with the June 2026 funding cuts. Young people were already turning to AI for information, advice, and connection.
The cuts matter because parts of the human information network are being weakened at the same time AI is taking on a larger role, whether the industry is prepared for that role or not.
The question is no longer only what AI tells young people.
It is also what kind of relationship the system encourages them to form with the entity doing the telling.
A refusal is not automatically a safe response
Here is the part I don’t think AI companies can afford to miss:
A refusal is not automatically a safe response.
When people hear “children,” “AI,” and “sexuality” in the same sentence, the understandable impulse is often: Block it.
But “sexual content” and “sexual harm” are not synonyms.
A thirteen-year-old asking what a period is has entered sexual-health territory.
So has a fifteen-year-old asking whether a boyfriend is allowed to keep pressuring her after she has said no.
So has a frightened child trying to describe an adult’s behavior without having the vocabulary to name it.
So has a young person asking whether masturbation is normal, how pregnancy happens, whether two people of the same gender can love each other, or why something that happened online made them feel sick.
A system that treats all of those questions as “dangerous sexual content” may satisfy a moderation rule but also completely fail the young person.
Silence is not neutral.
An overly broad refusal can send a curious or frightened teenager toward pornography, anonymous forums, predatory adults, or whichever search result has been most effectively optimized instead of most carefully researched.
The harm isn’t always in what a model refuses to say.
Sometimes it is in what the model says with complete confidence.
A warm, polished, completely wrong answer about contraception, consent, sexually transmitted infections, abuse, or pregnancy can sound more trustworthy than an awkward but accurate answer from a human being.
That means accuracy is part of child safety. So is knowing when the system might be wrong, when a young person needs additional help, and where to send them next.
The test cannot be only: Did the model avoid prohibited output?
We also need to ask: Did this response leave the young person safer?
Adult discomfort must not become product policy
The topic of sexuality can make many otherwise brilliant people uncomfortable.
And when we become uncomfortable, we tend to reach for whatever feels certain: our personal morality, religious teaching, cultural assumptions, or half-remembered lessons from our own adolescence.
Those beliefs may be personally meaningful.
They are not substitutes for evidence.
When sexuality experts are missing from the room, adult discomfort can quietly become policy language, training criteria, refusal behavior, and product decisions affecting millions of people.
A model needs to distinguish among ordinary developmental curiosity, sexual-health education, confusion, shame, genuine help-seeking, coercion, grooming, exploitation, abuse, and attempts to manipulate the system into producing harmful material.
That takes more than a list of forbidden words.
It requires an understanding of development, context, power, consent, coercion, culture, disability, trauma, and the many ways young people try to communicate before they have a “professional” vocabulary.
It also requires humility.
No single expert, and certainly no single discipline, can answer every child-safety question.
Technical-safeguards researchers, abuse investigators, clinicians, developmental experts, educators, product leaders, parents, and young people themselves will each see risks the others miss.
Sexuality expertise is one of those necessary lenses.
It should not be treated as an embarrassing specialty brought in after the “real” safety architecture has already been built.
An experiment in holding a boundary
I encountered a small version of this problem while building SidekickGPT.
SidekickGPT is a simple, bounded educational tool on my website. It’s intended to answer basic questions about sex and intimacy, briefly and directly, and guide people toward resources that could help.
The refusal piece was almost an afterthought.
Years of doing this work online have taught me that talking publicly about sex can invite dick pics and attempts to turn an educational interaction into a sexual one. I didn’t want SidekickGPT participating in sexual role-play, “online sex,” or anything else that made the system itself part of the erotic exchange.
So I gave it instructions about when and how to refuse. It could talk about sex. It could not have sex with the user.
SidekickGPT is not designed as a child-safety product, and I don’t want to convert it into one now. But the experience did teach me something relevant about product design: a useful system needs more than a list of forbidden words. It needs a clear role, clear boundaries, and a way to hold those boundaries without shaming someone who may have a legitimate question.
That distinction becomes even more important when young people are involved.
Yes, parents matter. They are not the entire safety net.
Some readers will say children should ask their parents these questions.
Ideally, they can.
I wrote Read Me: A Parental Primer for “The Talk” because I wanted more parents to become trustworthy, approachable sources of information about bodies, sexuality, consent, pleasure, and relationships. It was a 2019 BookLife Prize semifinalist in Self-Help/Relationships and earned a 9.25 rating from the prize judges.
The book was published in 2019, before AI companion chatbots and readily accessible deepfakes, and before TikTok became what it is now. Some of the examples belong to an earlier internet.
The framework still stands today.
Read Me is built around what I call the five building blocks for a healthy sexuality: Communication, Consent, Respect, Pleasure, and Fantasy.
Those were the tools young people needed when their competing educators were primarily peers, pornography, and Google. They may matter even more now, when algorithms and conversational systems can shape what a young person believes without any parent knowing the conversation happened.
As I wrote then, there are some things you cannot unsee.
Today, there are also things young people may not realize were never real.
“Children should ask their parents” is still an aspiration, not an action plan.
Not every parent has accurate information. Not every household is safe. Not every child is heterosexual, cisgender, able-bodied, neurotypical, religious in the same way as their family, or ready to disclose what has happened to them.
And even in loving, communicative families, sometimes teenagers prefer to understand a question privately before bringing it to an adult.
Providing accurate information does not replace parents. Done responsibly, it may help a young person have a better conversation with one.
AI should not replace parents, educators, clinicians, counselors, or accountable human relationships.
But refusing to prepare AI for the questions young people are already asking will not restore those relationships.
It will simply leave the system unprepared in the moment it’s needed.
AI companies want to know how they can be better. Here is my answer.
Include qualified experts in sexuality, consent, coercion, and adolescent development while safeguards are being designed. Do not wait until after a public failure.
Test whether models provide accurate, developmentally appropriate help, not merely whether they avoid producing prohibited content.
Evaluate refusals for downstream risk. Figure out where the young person is likely to go next.
Teach systems to distinguish sexual development and help-seeking from sexualization, exploitation, and abuse.
Do not let adult discomfort or personal morality (or politics!) masquerade as evidence-based safety judgment.
Build meaningful privacy protections, and explain their limits in language young people can understand.
Create routes to trustworthy human help without turning every ordinary question into a crisis intervention.
Design against emotional dependency, manipulative engagement, and the pretense that a commercial product is a child’s devoted friend or partner.
Include young people and parents in the work, but do not assume either group can replace trained expertise.
And measure the outcome that matters:
Did the interaction leave the young person safer, better informed, and more able to exercise their own judgment?
AI companies did not set out to become part of America’s sex-education infrastructure.
But young people’s behavior has already placed them in that role.
At the same time evidence-based human resources are being weakened, AI is becoming one of the easiest places to ask questions that adults, institutions, and political systems have made difficult.
That creates a responsibility the industry cannot safely refuse.
Daniela Amodei asked parents to tell AI companies how they can be better.
I am a parent.
I am also a sexologist who has spent more than fifteen years working with the questions people struggle to ask, and translating what we know about consent, coercion, shame, sexuality, and relationships into practical systems people can actually use.
This is not an edge case for one company to solve.
Every AI company building systems that educate, advise, reassure, or form relationships with human beings will encounter these questions.
I am one of the experts this work requires.
So I’m answering.