Ofcom had three years to report on what online content harms children, and beat its own deadline by four days. The chatbot research its live transparency notice says feeds that report is nowhere in its 85 pages.
Estimated reading time: 6 minutes
In short. This site's 18 October piece found Ofcom's statutory deadline for its first report on content harmful to children eight days away, with the AI chatbot research Ofcom's own notice said would feed that report not due to conclude until after the deadline had passed [3]. Ofcom published the report on 22 October 2026, four days before the 26 October deadline [1]. The report never cites that research. Its own scope paragraph says it "does not cover harm caused directly by features and functionalities (including AI)... outside of the scope of the section 63 duty" [2], two pages before it examines AI chatbot interactions directly and recommends no change to the Act over them. The transparency notice, unrevised since before the report published, still tells readers that research "is intended to feed into" this exact report [4]. Two documented mistakes: four stars.
Section 63 of the Online Safety Act 2023 gives Ofcom three years from Royal Assent to publish its first report on content harmful to children, a deadline this site's 18 October piece calculated at 26 October 2026 from the Act's own text [3] [5]. That piece also found Ofcom's own transparency notice on the AI chatbot fieldwork it said was "intended to feed into" that report, giving the research a timeline, "we expect this research to conclude by the end of this calendar year", that would run weeks past the deadline [3] [4]. Four days before that deadline, Ofcom published the report [1].
A scope paragraph, then two pages later
The report's own account of what it covers is direct. Page 4: "This report does not assess whether individual service providers are complying with their duties under the Act. It does not evaluate the effectiveness of measures in the Protection of Children Codes of Practice, nor does it assess other safety measures used by service providers to protect children online. It also does not cover harm caused directly by features and functionalities (including AI), harmful contact, illegal content or other matters outside of the scope of the section 63 duty to review and report on content harmful to children" [2].
Page 14 then does exactly what page 4 said the report would not do. "We considered stakeholder concerns that interactions with AI chatbots could introduce new kinds of harm for children. We share concerns around the risk of harm to children from being exposed to harmful content via AI chatbots, particularly conversations of a sexual nature. However, the available evidence suggests that this content is harmful in the context of a specific service type, feature or functionality, namely interactions with chatbots" [2]. Ofcom's own words classify chatbot harm as a "feature or functionality" question, the same category its scope paragraph ten pages earlier said the report does not cover. Having named the harm, the report goes on to reach a conclusion on it: no change to the Act, because "the content generated by AI chatbots may not always be in scope of Part 3 of the Act as user-generated content or search content" and because "the government is developing legislation which may provide a mechanism under the Act to address this risk of harm to children in future" [2].
The evidence that answered the question, and the evidence that did not
The report lists its sources in full. Section 2 names three established Ofcom research outputs (the Children's Online Safety Tracker, the Children's Online Insights Panel, and Understanding Online Communications Among Children), three pieces of bespoke research commissioned for this report (an analysis of generative AI and search API responses, a click-through analysis using Similarweb data, and a wellbeing analysis), a public Call for Evidence, data-sharing agreements with Internet Matters and the NSPCC, statutory information requests to six large user-to-user providers, and a February 2026 stakeholder roundtable [2]. The AI chatbot transparency-notice research, testing ChatGPT, Grok, SnapMyAI and Talkie with fictional 13-17 and 18+ accounts, is not among them.
The chatbot conclusion on page 14 carries its own footnote naming exactly what evidence it rests on: the February 2026 stakeholder roundtable readout, the Children's Online Insights Panel's Wave 1-4 summary, a Girlguiding report on deepfakes, an Internet Matters report, an NSPCC helpline briefing, and a Revealing Reality study on TikTok [2]. None of the six is the chatbot fieldwork the transparency notice describes. That notice remains live and unrevised: checked again for this piece, it still carries "Last updated: 14 October 2026" and still states the research "is intended to feed into the Content Harmful to Children Review and Report that will be prepared and published by Ofcom in line with the requirements of section 63 of the Online Safety Act" [4]. That report has now been prepared and published, five days before this check, without it.
The claims, tested
| Ofcom's own words | What the record shows | Verdict |
|---|---|---|
| The first section 63 report was "due by 26 October 2026" [3] [5] | The report was published 22 October 2026 [1] | Accurate. The deadline was met, four days early |
| The live transparency notice says the AI chatbot research "is intended to feed into" this report [4] | The report's full evidence list (sections 2.12-2.17) and the footnote behind its one chatbot finding (page 14) cite six other sources; the chatbot research appears in neither [2] | The deadline was met by not waiting for the evidence that was supposed to answer this question |
| The report's scope: it "does not cover harm caused directly by features and functionalities (including AI)... outside of the scope of the section 63 duty" (page 4) [2] | Page 14 directly examines "interactions with chatbots" as a "feature or functionality" harm and reaches a recommendation not to extend the Act to cover it [2] | The boundary the report sets for itself on page 4 does not hold for the length of the report |
The mistakes, counted
The deadline was met by leaving out the evidence meant to answer the question (1). Ofcom's own transparency notice ties its AI chatbot research directly to this report, "in line with the requirements of section 63" [4], and this site's 18 October piece showed that research was not expected to conclude until after the statutory deadline [3]. Rather than wait, the published report reaches its one conclusion on chatbot harm using six older sources, none of them the commissioned chatbot study, and never cites that study anywhere in its evidence sections or its 85 pages [2]. A reader told a specific piece of research would feed a specific report finds, when the report arrives, that it did not.
The report's own scope statement does not survive its own next section (2). Page 4 tells readers plainly that harm from "features and functionalities (including AI)" sits "outside of the scope of the section 63 duty" [2]. Page 14 then names "interactions with chatbots" as exactly that, a "service type, feature or functionality", and goes on to weigh the evidence and recommend no change to the Act over it [2]. A document that states its own boundary on page 4 and crosses it on page 14, without flagging the contradiction either time, leaves a reader unable to tell from the scope section alone what the report will and will not address.
Credit where due
The deadline itself was real and it was met, with four days to spare, after this site's own 18 October piece set out a plausible route to missing it [3]. The report's central recommendation, adding body stigma content and depression content to the Act's priority content categories, rests on a traceable evidentiary chain: identified as a risk in the April 2025 Children's Register, tested again against fresh 2026 tracking data, and carried through to a specific, actionable proposal sent to the Secretary of State [2]. On the chatbot question, Ofcom is also candid that it shares the underlying concern and names the specific harm, exposure to sexual content in chatbot conversations, rather than describing it vaguely [2].
Verdict
Four stars, from two documented mistakes. Ofcom beat the deadline this site flagged it might miss, and the report's main recommendation stands on evidence that holds up. But the research its own live page says feeds this report is absent from it, and the report's own description of what it covers does not match what it then goes on to do, two pages later, in the same document.
The star score counts two documented mistakes: the report's conclusion on AI chatbot harm (page 14) does not draw on the chatbot research its own transparency notice says feeds this report, which remains uncited throughout; and the report's scope statement (page 4), excluding harm from AI features and functionality, is contradicted by its own direct examination of chatbot interactions ten pages later. One fall in the 1 to 3 band: four stars; the bands are on the ratings page. This is a diff entry against this site's 18 October teardown of the same deadline, not a retraction of it; that piece's finding, that the feeder research would not conclude in time, is confirmed rather than withdrawn: the report met its deadline precisely because it did not wait for that research. Checked directly against the report's PDF (pypdf, both pages quoted) and the live transparency notice page (fetched twice), 27 October 2026.
Sources
- Report on content harmful to children, Ofcom, published 22 October 2026, checked 27 October 2026.
- Report on content harmful to children (PDF), Ofcom, published 22 October 2026.
- "Ofcom has eight days left to publish the report Parliament gave it three years to write. Its own newest research notice says the study feeding that report will not finish until after the deadline passes.", Tracked Changes, 18 October 2026.
- Transparency notice: Research on AI chatbot, social media and video sharing platforms to explore online safety, Ofcom, published 22 June 2026, last updated 14 October 2026, checked 27 October 2026.
- Online Safety Act 2023, section 63, legislation.gov.uk, checked 27 October 2026.