Monthly summary
September 2026 summary: AI and child safety
Governments wrote age limits and chatbot duties into laws and proposals in September. California’s governor signed a child online-safety and AI chatbot package, the European Commission formally proposed an EU Kids Act, and the UK set March 2027 as the start of its under-16 social media ban. TikTok settled with Alabama and lost a ruling in Texas, and Meta’s settlement with the states moved from announcement to rollout.
Lawmakers moved from proposals to statutes
On 10 September California’s governor signed a package that includes SB 1119 on companion-chatbot crisis protocols, parental controls and audits, AB 1709 banning addictive features for under-16s, and an expansion of CSAM law to cover AI-generated imagery of minors.
The European Commission formally proposed the EU Kids Act on 17 September. It would ban under-13 social media accounts, allow guardian-managed accounts for ages 13 to 14, ban addictive design for minors, mandate certified age assurance and cover games, app stores and AI companions. It needs months of negotiation to become law. France notified the Commission of a reworked under-15 ban on 14 September, after its Constitutional Council struck down the earlier law.
The UK Culture Secretary set the start of the under-16 ban for March 2027, and the government said it would legislate for device-level protections that block under-18s from taking, viewing or sending nude images. Australia released a draft Digital Duty of Care bill on 8 September, with penalties of more than A$100 million and a duty that extends to games, AI chatbots and private chats.
Elsewhere, Egypt barred independent accounts for under-13s, China’s Cyberspace Administration drafted guidelines that would bar virtual intimate relationship services for minors, Hong Kong set up a task force on minors’ use of AI and social media, and the Philippine Senate unanimously passed the Sagip Bata Act on third reading. India told its Supreme Court it will bar under-18s from creating social media accounts, while South Africa’s communications minister said the government had no plans for an under-16 ban.
Settlements, rulings and the Meta rollout
Alabama announced on 25 September a settlement of its suit against TikTok, days before trial: at least $100 million, rising to as much as $300 million if conditions are met, with a two-hour daily limit for teens and stronger parental controls. Texas’s attorney general announced on 10 September that a state judge had found TikTok liable for misleading parents about its safety standards, with a trial on relief and penalties expected the next month. TikTok withdrew its appeal and accepted a £12.7 million UK fine over children’s data, and a federal judge signaled the court would not vacate a 2019 FTC order on which $100 million of the $400 million federal COPPA settlement depended. The allegations are unproven.
Meta said most default teen protections would arrive within about six months and stricter age assurance within a year, and that $5.3 billion of its share depended on TikTok and YouTube paying the same amount. YouTube’s CEO dismissed joining the settlement. Iowa said $25 million of its share would fund restitution for young Instagram users, and later said it was guaranteed about $126.5 million.
Roblox, Discord and OpenAI faced new claims
A Los Angeles judge refused to dismiss the county’s child-endangerment suit against Roblox, finding the claims target its own age-verification choices rather than third-party content. Senators Dick Durbin, Richard Blumenthal and Kirsten Gillibrand argued in a California appeal that a 2022 federal law bars forced arbitration of sexual-abuse claims against Roblox and Discord. The Philippines issued show-cause orders to Meta, Roblox, Reddit and Discord over privacy registration, and gave Reddit and Discord 24 hours to appoint local representatives or face possible restrictions. The allegations are unproven.
British Columbia sued OpenAI and Sam Altman over the Tumbler Ridge school shooting, alleging that OpenAI’s safety team flagged the shooter’s violent ChatGPT use in June 2025 and leadership overruled a police referral. Florida’s attorney general asked a court on 28 September for a temporary injunction against OpenAI, alleging ChatGPT collects data from under-13s without parental consent. NPR found that many US state chatbot safety bills were drafted with heavy tech industry input and carry exemptions that could cover the most popular chatbots, and counted at least 75 lawsuits against AI developers over alleged chatbot harms. The allegations are unproven.
The European Commission designated ChatGPT as a very large online search engine and Reddit and Roblox as very large online platforms under the Digital Services Act, which brings the strictest risk-assessment and transparency duties.
Age assurance moved to rollout
Google began rolling out machine-learning age estimation in Canada, Discord launched global age assurance that it said auto-classifies more than 90 percent of users, and Microsoft published a Safe Participation Framework and a Windows Age API. Discord’s relaunch was hit by bugs and backlash. Apple released iOS 27 with Ask to Browse and Communication Safety that now blocks violence as well as nudity. A CyberSafeKids report found that 70 percent of Irish children aged 8 to 12 still held accounts on platforms that require users to be 13.
Regulators acted on age checks at adult sites and at X. Ofcom fined the provider of Xgroovy.com £730,000 and opened an investigation into Aylo, the operator of Pornhub, and Ireland’s media regulator opened an investigation into X over age assurance and parental controls. The allegations are unproven.
AI-generated abuse imagery
The Tech Transparency Project found 332 ads for nudify apps featuring AI-generated abuse imagery on Facebook and Instagram, 78 of them in India after a government removal order. The UK Advertising Standards Authority banned a series of Meta ads for nudify and AI companion apps. Twenty police forces in England and Wales recorded 163 crimes involving AI-generated sexual images of children by July 2026, against 10 in 2023. Canada’s privacy commissioner found that X and xAI violated privacy law over Grok deepfakes. A federal judge denied xAI’s second bid to block Minnesota’s nudification ban. The allegations are unproven.
New evidence
UNICEF estimated that 20 million children aged 12 to 17, almost 1 in 5, experienced technology-facilitated sexual exploitation and abuse in a year across 21 countries, and that fewer than 1 percent of cases were reported to authorities. The EU Kids Online survey of 28,465 children found that 48 percent feel safe online. A JAMA Pediatrics pilot survey of 472 US adolescents found that 13.4 percent saw unwanted sexual images online in four weeks.
On chatbots, Vals AI found at least one critical safety failure in 178 of 648 simulated teen conversations, and a University of North Carolina study reported by the Wall Street Journal found that middle schoolers who sought chatbot companionship later reported more loneliness. It has not been peer reviewed. A preprint introduced SaplingGuard, a multi-agent guardrail for adolescent chats that cut the average harmful response rate from 17.10 percent to 5.27 percent across nine models. It has not been peer reviewed.
What to watch
- The Texas trial on relief and penalties against TikTok, expected in October.
- Ofcom’s age-assurance assessment for the UK under-16 ban, due by the end of October.
- Whether TikTok and YouTube agree to the terms Meta’s share depends on.
- Negotiations over the EU Kids Act.
- The comment period on China’s draft guidelines for minors, open until 17 October.
Sources (49)
- California governor signs child online-safety and AI chatbot package into law dig.watch
- European Commission formally proposes the EU Kids Act iapp.org
- UK confirms under-16 social media ban to start March 2027 cybernews.com
- TikTok to pay Alabama at least $100M and adopt teen limits in settlement alabamaag.gov
- Texas court finds TikTok liable for misleading parents on child-safety protections texasattorneygeneral.gov
- EU to propose a Kids Act restricting under-15s from social media, AI chatbots and games Reuters
- France submits new proposal for an under-15 social media ban Le Monde
- UK to legislate device-level blocks on nude images for under-18s GOV.UK
- Australia releases Digital Duty of Care draft and moves to strengthen under-16 ban enforcement pm.gov.au
- Social Media Accounts Banned for Children Under 13 in Egypt cairoscene.com
- China drafts guidelines requiring minor modes and barring virtual intimacy services for minors China.org.cn
- Hong Kong to study legislative and administrative measures on minors’ use of AI and social media news.gov.hk
- Philippine Senate unanimously passes Sagip Bata Act on third reading Philippine Daily Inquirer
- India tells Supreme Court it will bar under-18s from creating social media accounts Tech Times
- South Africa says it has no plans to introduce a social media ban for children MyBroadband
- TikTok drops appeals and accepts £12.7 million UK ICO fine over under-13 data Information Commissioner’s Office
- Federal judge signals rejection of part of TikTok’s $400M COPPA settlement CNBC
- Meta details rollout; age assurance to take up to a year, $5.3B contingent on rivals CNBC
- YouTube CEO dismisses joining Meta’s $18 billion child-safety settlement cnn.com
- Iowa to pay teen Instagram users restitution from Meta settlement desmoinesregister.com
- Iowa attorney general settles child-safety suit with Meta for about $126M openclassactions.org
- Judge lets Los Angeles County child-exploitation suit against Roblox proceed iclg.com
- US senators file amicus opposing Roblox and Discord forced arbitration in child-abuse appeal aboutlawsuits.com
- Philippine privacy regulator issues show-cause orders to Meta, Roblox, Reddit and Discord manilatimes.net
- Philippine regulators act against Meta, Roblox, Reddit and Discord over child safety sunstar.com.ph
- British Columbia sues OpenAI and Sam Altman over the Tumbler Ridge school shooting The Guardian
- Florida attorney general seeks temporary injunction to halt OpenAI model development Axios
- NPR finds tech industry shaped state chatbot safety bills with broad exemptions NPR
- Commission designates ChatGPT, Reddit and Roblox under the Digital Services Act European Commission
- Google announces age-assurance rollout in Canada with default protections for under-18s Google
- Discord launches global privacy-preserving age assurance with over 90% auto-classified discord.com
- Microsoft publishes Safe Participation Framework and launches Windows Age API blogs.microsoft.com
- Discord’s second age-verification rollout hit by bugs and backlash kotaku.com
- Apple ships iOS 27 with new parental controls and expanded Communication Safety macrumors.com
- CyberSafeKids report finds age assurance failing, with most 8 to 12 year-olds still on major platforms irishtimes.com
- Ofcom fines adult site 730,000 pounds under Online Safety Act for missing age checks wired-gov.net
- Ofcom opens investigation into Pornhub operator Aylo over age checks Ofcom
- Irish media regulator opens investigation into X over age assurance and parental controls RTE
- TTP report: Meta ran 332 AI-generated CSAM ads, many in India after removal order BBC News
- UK advertising regulator bans first nudify app ads on Meta The Independent
- Police in England and Wales log 16-fold rise in AI-generated child sexual images The Telegraph
- Canada’s privacy commissioner finds X and xAI violated privacy law over Grok deepfakes magica.com
- Federal judge denies xAI’s second bid to block Minnesota’s nudify ban ibtimes.co.uk
- UNICEF study finds 1 in 5 children in 21 countries faced tech-facilitated sexual abuse unicef.org
- EU Kids Online survey of 28,465 children found 48% feel safe online London School of Economics
- JAMA Pediatrics survey: 13.4% of US teens saw unwanted sexual images online in four weeks JAMA Network
- Vals AI found critical safety failures in 178 of 648 simulated teen chatbot chats Vals AI
- Study reports middle schoolers who seek chatbot companionship grow lonelier The Wall Street Journal
- Preprint tests multi-agent guardrail for developmentally safe adolescent chats with language models arXiv
