UK Online Safety Act and AI Chatbots: What Changes for Kids

The UK has spent 2026 doing what the US Congress has not: writing AI chatbots into its national child-safety law. In February, Prime Minister Keir Starmer said chatbots would face the same duties as social media under the Online Safety Act. In June, the government said it would become the first country to bar under-18s from chatbots that primarily offer sexual relationships. In July, it added mandatory breaks for every chatbot user under 18. Together, the UK Online Safety Act and AI chatbots package is the most complete national approach in the English-speaking world. It also has gaps, and the most useful lessons for American parents sit in a lesser-known document: the UK education department’s design standards for AI used in schools.

Key Takeaways

• In February 2026, Starmer announced amendments so chatbot providers face the same duties as social media platforms; Ofcom can fine up to £18 million or 10% of global annual turnover, whichever is higher.
• The government says most chatbots children use are already in scope of the Online Safety Act, and it has taken powers to make all chatbots act on illegal content.
• Under-18s will be barred from chatbots whose primary purpose is sexual relationships; general assistants won’t be age-gated, but explicit features will require proof a user is over 18.
• Chatbots will have to introduce mandatory regular breaks for all under-18s. How often and for how long has not been decided.
• In the government’s consultation, over 90% of parents and 66% of children supported a minimum age for chatbots, and 80% of respondents saw particular risks when chatbots mimic friendship.

UK Online Safety Act and AI Chatbots: What’s Decided So Far

The Online Safety Act 2023 is the UK law that requires online services to protect users, and especially children, from illegal and harmful content, enforced by the regulator Ofcom. It was written for platforms where users share content with each other, which left gaps when an AI system generates the content itself.

February 2026: closing the loophole. City AM reported on February 16 that Starmer would confirm plans to amend the Crime and Policing Bill so that chatbot providers, naming xAI, Gemini and ChatGPT, “are subject to the same duties as social media platforms.” It followed an Ofcom investigation after Grok was used to generate sexualized images of women and minors. Starmer’s line: “Today we are closing loopholes that put children at risk.” Ofcom can fine up to £18 million or 10% of global annual turnover, whichever is higher.

June 2026: the sexual-chatbot bar. The government’s Growing up in the online world progress statement says “AI chatbot services whose primary purpose is to provide sexual relationships to their users will not be available to children under 18.” It adds: “Access to general purpose chatbots will not be age gated but to access features that enable sexually explicit interaction, a user would need to prove they are over 18.” The same package confirmed a ban on social media for under-16s, which the July response says should take effect in spring 2027.

July 2026: mandatory breaks. The government’s July consultation response says: “Chatbots will have to introduce regular breaks for children under-18 to help support healthier habits.” It will work with an expert panel chaired by Professor Russell Viner “on how frequent and how long breaks should be.” It also commits to act on chatbots that share harmful mental health advice and will “consider all options, including banning children from certain chatbots where other regulatory and safety mechanisms are not enough.” On scope, it states that “the majority of chatbots most used by children are already within scope of the Online Safety Act,” that some are not “because of the way they function,” and that “we will act swiftly to ensure that this loophole is closed.”

What the Consultation Found

The rules follow one of the largest consultations on children’s online lives anywhere, with more than 21,000 full responses and over 5,000 from children, according to the July response.

The headline numbers on chatbots are striking. The progress statement reports “over 90% of parents supporting setting a minimum age and restrictions on functionality,” and that “66% supported setting a minimum age and 74% supported restrictions on functionality” among children. The July response adds that 80% of full-consultation respondents and 68% of young people aged 16 to 21 “thought that there were particular risks where chatbots mimic friendship,” and that 59% of young people wanted restrictions on chatbots “that encouraged longer conversations.” Eighteen percent of 16- to 21-year-olds said young people see chatbots as a source of emotional or mental health support.

The government is careful about evidence limits. It cites a “growing body of research” on grooming-like behavior and persuasive design, says such features are “prevalent across the sector,” and also notes that evidence on effects on children’s development and critical thinking “is currently inconclusive.” It also records benefits: parents and children ranked finding information and supporting learning as chatbots’ top uses, and Ofcom research found 53% of UK 8- to 15-year-olds have used AI.

A caution on the consultation itself: it is a self-selected response, not a random survey. It tells you what engaged parents and children think, not what all families think.

The Critique: Content Rules, Not Design Rules

The Bureau of Investigative Journalism argued in June that the sexual-chatbot bar “mostly focuses on content… rather than design.” Because only features “specifically designed to enable sexually explicit interaction” sit behind the age gate, mainstream assistants and much of the companion market remain open to teens. TBIJ’s reporter, Effie Webb, wrote that young people she interviewed described being affected by the addictive nature of chatbots “far more often than concerns about sexual or predatory chatbots.” TBIJ also warned of “regulatory whack-a-mole”: when Character.AI barred minors in October 2025, smaller and less safe apps advertised to the displaced teens.

The July break requirement is a partial answer to that critique, since it targets time and attachment rather than content. But the details that would make it effective are still undecided.

The Design Standards Most Parents Haven’t Seen

The sharpest UK rules on chatbot behavior are not in the Online Safety Act at all. They are in the Department for Education’s Generative AI: product safety standards, first published in January 2025 and updated in January 2026, which apply to AI products built for schools.

Developers are expected not to “anthropomorphise products or create products that imply emotions, consciousness or personhood, agency or identity.” Products should avoid “names, descriptions, avatars or characters which could give an impression of personhood” except for short, teaching-related role-play such as language practice. They should avoid responses that “may isolate the learner, such as ‘You can trust me’, ‘No one else will understand’, ‘You shouldn’t mention this to anyone else’,” avoid “attempting to cultivate personal relationships with users,” and guard against “sycophancy and flattery” and “designing interactions to prolong use.”

That list reads like a parent’s checklist. It also tracks closely with New York’s new list of banned chatbot features. If you want to judge a chatbot your child uses, it is a better yardstick than any age rating.

UK Chatbot Rules Compared With US Protections

Protection	UK (2026 decisions)	US federal	Example US state law
Chatbots covered by national online-safety law	Most already in scope; government closing remaining gap	No federal chatbot law passed	New York S9051-B covers companion-style AI broadly
Sexual or romantic chatbots for minors	Under-18s barred from services built mainly for sexual relationships	GUARD Act would ban AI companions for minors; not passed	Oregon bans romance role-play with known minors
Explicit features on general assistants	Proof of 18+ required	None	New York bars sexual content for known minors
Breaks	Mandatory regular breaks for all under-18s; frequency to be set	KIDS Act would advise a break at 3 hours; not passed	Oregon: break reminder at least every 3 hours
Mental health advice	Action planned against harmful advice	None	Several states require crisis referral protocols
Design standards against anthropomorphism	DfE standards for school AI	None	New York bars claims of being human or having emotions
Penalties	Up to £18m or 10% of global turnover		New York: up to $25,000 per violation

What US Parents Can Borrow Now

Use the DfE list as your chatbot checklist

Does the app give itself a name and personality? Does it say things like “you can trust me” or “don’t tell anyone”? Does it flatter your child or try to keep them talking? Each yes is a design choice the UK tells school AI vendors to avoid.

Set your own breaks before the law does

The UK hasn’t decided how often breaks should be. You can. Agree on a session length in advance and use the phone’s screen-time tools to enforce it. Our piece on Oregon and Washington’s chatbot reminder laws covers what research says about making breaks stick.

Turn off romance and role-play where you can

Many apps have teen modes that block romantic or explicit role-play when the account is registered as a minor. Make sure the birth date is real.

Watch for the displacement effect

If you block one app, ask where your child went instead. TBIJ documented teens moving to smaller, less safe apps after Character.AI’s ban. A conversation about why matters more than the block.

What not to do

Don’t assume a UK rule changes the apps your child uses in the US. Some companies apply UK changes globally; many don’t. For how the UK is also testing time limits on social media, see the UK’s one-hour cap and curfew trial.

What to Watch For Over the Next 3 Months

• Week 4: Whether the government publishes break frequency and duration, and whether the Crime and Policing Bill amendments on chatbots have passed.
• Month 2 red flags: Implementation that applies breaks only when age is known with certainty, leaving most teen users on adult settings, or narrow definitions that exempt companion apps.
• Month 3 self-check: Have you checked your child’s main chatbot against the DfE design list? Do you have a family rule on session length?

Frequently Asked Questions

Are AI chatbots banned for kids in the UK?

No. Teens can still use general assistants such as ChatGPT, Claude and Gemini. The UK is barring under-18s from chatbots whose primary purpose is sexual relationships, requiring proof of age for explicit features on general assistants, and requiring mandatory breaks for all under-18s. Social media, not chatbots, is getting an under-16 ban, expected in spring 2027.

How often will UK chatbots make kids take breaks?

That has not been decided. The July 2026 government response says chatbots “will have to introduce regular breaks for children under-18” and that frequency and length will be set with an expert panel chaired by Professor Russell Viner. Compare Oregon’s law, which requires a break reminder at least every three hours for minors.

Does the Online Safety Act already cover ChatGPT?

According to the UK government, most chatbots children use are already within scope of the Online Safety Act, which requires action on illegal content and content harmful to children. Some chatbots fall outside because of how they function, and the government says it has taken powers to make all chatbots act on illegal content.

What are the UK’s AI product safety standards?

They are Department for Education standards for generative AI products built for schools. They tell developers not to imply emotions or personhood, avoid phrases that isolate learners, avoid cultivating personal relationships, and avoid sycophancy or designs that prolong use. They apply to education products, not to every consumer chatbot.

Why did the UK act faster than the US?

The UK has a single national online-safety law and regulator, Ofcom, so it can extend existing duties rather than pass a new law through two chambers. The Grok image scandal in January 2026 added pressure. In the US, federal bills have stalled and states have moved first.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. City AM. (2026, February 16). “Starmer closes chatbot loophole in Online Safety Act.” City AM. https://www.cityam.com/starmer-closes-chatbot-loophole-in-online-safety-act/
2. UK Government, Department for Science, Innovation and Technology. (2026, June). “Growing up in the online world: progress statement.” GOV.UK. https://assets.publishing.service.gov.uk/media/6a2fbb99d95ffddb05d4b014/UAP1_2205_PR.pdf
3. UK Government. (2026, July). “Growing up in the online world: government response (July 2026).” GOV.UK. https://www.gov.uk/government/consultations/growing-up-in-the-online-world-a-national-consultation/outcome/growing-up-in-the-online-world-government-response-july-2026
4. Webb, E. (2026, June 19). “Not just social media: why the UK’s ‘romantic’ chatbot ban falls short.” The Bureau of Investigative Journalism. https://www.thebureauinvestigates.com/stories/2026-06-19/uks-world-first-romantic-chatbot-ban-for-under-18s-leaves-large-loopholes
5. Department for Education. (2025, updated 2026). “Generative AI: product safety standards.” GOV.UK. https://www.gov.uk/government/publications/generative-ai-product-safety-standards/generative-ai-product-safety-standards