Ofcom 拟将身材羞辱与抑郁内容列为儿童保护优先内容

The UK’s Office of Communications (Ofcom) has recommended that platforms be required to protect children from body-shaming content and content that promotes depression. However, its own evidence cannot yet show whether its existing child safety rules work.

In its first report on content harmful to children, published on October 22, Ofcom found no significant change in children’s exposure to harmful content in the nine months after its rules took effect in July 2025. By then, YouTube and TikTok had still not committed to the main change Ofcom had called for. The two new categories also depend heavily on context, which automated systems struggle to interpret, and the report does not estimate the cost of enforcing the proposed rules.

The report does not claim to evaluate the rules. “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 said.

What did Ofcom find?

Most children still see harmful content. The following figures show the share of 11-to-17-year-olds who saw each type of content in the four weeks before Ofcom’s March-April 2026 survey:

Dangerous stunts or challenges: 29%

Feeds: Children most often encountered harmful content through personalised feeds and, to a lesser extent, through comment sections.

Search: Search engines were the entry point for 30% of children’s visits to pornography sites. In Ofcom’s tests, both generative AI-powered search tools and traditional search engines returned links to pornography sites without age checks for 20.3% of queries.

Vulnerable children: Children with health conditions or special educational needs were more likely to encounter depression and body stigma content. Girls, non-white children and children with limiting conditions were more likely to encounter body-shaming content.

Wellbeing: Children who saw these types of content reported lower wellbeing. Ofcom cautioned that its results “should not be treated as causal.”

What has Ofcom recommended?

Ofcom wants body stigma and depression content added to the Online Safety Act’s list of “priority content”. It defines the two categories as follows:

Body stigma content: Content “that shames or otherwise stigmatises body types or physical features.”

Depression content: Content “that promotes depression, hopelessness and despair,” including content that “glamourises, glorifies or romanticises” them.

Both categories currently count as “non-designated content”. Platforms must address content in this tier only if their own risk assessments flag it. If the categories become priority content, every platform within the scope of the rules would have to protect children in at-risk age groups from it. Ofcom said the evidence of harm, particularly from repeated exposure, is now “sufficiently established” to justify the change.

Ofcom decided against adding two other categories:

AI chatbots: Ofcom said the harm stems from a type of service rather than a type of content. The government is also drafting separate legislation on the issue.

Graphic, extremely violent content: Ofcom considered prohibiting such content for all children, citing cases of children seeing footage of Charlie Kirk’s death. However, it held off because the government plans to criminalise the most extreme forms of such material.

What did Ofcom ask platforms to do in 2025?

Ofcom’s Protection of Children Codes took effect on July 25, 2025. They contain more than 40 measures, divided into two broad groups.

Design measures change how a platform works:

Safer feeds: Platforms at medium or high risk must keep harmful content out of children’s feeds or give it “a low degree of prominence”. Since feeds are the main route through which children encounter harmful content, this measure targets a major source of exposure.

Age checks: The riskiest services must use “highly effective age assurance”, such as ID checks or facial age estimation.

Controls: Children must be able to block and mute accounts, turn off comments and decline group chat invitations.

Governance: Every service must appoint a named person accountable for children’s safety.

Content measures depend on assessing individual posts:

Moderation: All services must act quickly on harmful content once they become aware of it.

Support: Children who search for harmful content must be shown supportive information.

The new recommendation would expand the content measures by widening the range of content platforms must assess.

Did platforms follow the rules?

Platforms widely adopted age checks, but the feed-related requirements remain disputed:

Age checks: All 10 of the most popular pornography sites in the UK now check users’ ages, according to Ofcom’s July report on age assurance. The share of children asked to prove their age rose from 41% to 51%.

Feeds: In May 2026, Ofcom said YouTube and TikTok had not “committed to any significant changes in response to our specific demands” on feeds, ITV reported. Both companies said their feeds were already safe for children.

Moderation: Data from seven major platforms, including Instagram, TikTok and YouTube, showed no consistent trend. Some platforms acted on more harmful content in 2025, while others acted on less.

Is there evidence that the rules worked?

Evidence of the rules’ effects is limited to measures that platforms have clearly adopted:

Age checks: “Highly effective age assurance is helping to prevent children from accessing pornography services,” Ofcom said.

Feeds: The share of children who said they encountered harmful content through feeds fell from 37% before the rules took effect to 35% afterwards.

Overall exposure: About seven in 10 children still reported seeing harmful content, a proportion similar to that recorded before July 2025.

Ofcom’s own guidance cautions against judging the rules too early. Its 2024 paper on evaluating safety measures said that “the impacts of a safety measure could take time to be realised”. It also said regulators need a baseline against which to measure changes. The October report has three gaps when assessed against that standard:

No baseline: Ofcom first asked children about body stigma and depression content in late 2025, after the rules took effect.

No platform data: The moderation data Ofcom collected does not cover these two categories.

No breakdown: The survey cannot separate the effects of age checks from those of feeds. Nor can it distinguish between platforms that changed their practices and those that did not.

Can platforms identify body-shaming and depression content?

Both categories depend heavily on context. A sad post does not qualify as depression content unless it presents despair as appealing. Similarly, a comment about someone’s body could be a joke or an instance of harassment. Research suggests that automated systems can identify such content only partly accurately:

Context: “It can sometimes be difficult for our systems to distinguish between a bullying comment and a light-hearted joke,” Meta said in 2021. At the time, its systems detected 59.4% of the bullying content that Facebook removed before users reported it.

False positives: A 2017 study of a classifier designed to identify content promoting eating disorders found that only about half of the posts it flagged violated the rules.

Humour: A 2025 study on body-shaming comments found that such remarks are often “disguised as humor”.

Video: In a 2025 test involving 1,500 videos, human reviewers were more accurate than every AI model tested. The models also performed worse on non-English content.

There are signs that the task is at least partly manageable. YouTube already limits repeated recommendations to teenagers of videos that idealise certain body types. Ofcom’s codes also generally require platforms to reduce the prominence of such content in feeds rather than remove it altogether. An incorrect decision about visibility therefore carries a lower risk than an incorrect removal.

The risks, however, run both ways. In January 2025, Meta said it would rely on user reports for “less severe policy violations”. After the codes took effect, Reddit blocked access in the UK to support communities focused on addiction and sexual assault. If platforms mistakenly suppress recovery-related content, they could harm the very children the rules aim to protect.

Ofcom has not specified the level of accuracy it expects. In June 2025, it proposed that larger platforms use automated detection where it is “technically feasible”. It called for “an appropriate balance between precision and recall” but did not set a numerical standard, according to a legal analysis.

What would enforcement cost?

The report does not estimate the cost of implementing its recommendation. Ofcom’s earlier figures provide some indication of the potential expense:

Moderators: Ofcom estimates that a moderator costs between £38,430 and £76,860 a year.

Images: Ofcom estimated that developing a tool to match known illegal images costs between £17,000 and £339,000. Assessing content that depends on context is a more difficult task.

Video: In the 2025 test, human review cost about $0.65 per video, compared with $0.02 to $0.04 for the cheapest AI models.

Audio: Voice chat presents an even greater challenge. “Voice channels just basically can’t be moderated,” a Discord moderator told researchers in a 2019 study, because conversations are usually not recorded.

Whether these costs favour large platforms is an open question. Ofcom’s heaviest measures apply only to services with over seven million UK users or several kinds of risk, which shields small services. Every service would still have to assess the new categories, though, and large platforms can spread the cost of classifiers and moderators across far more users. In 2024, a small UK cycling forum said it would close because compliance would cost “likely tens of thousands” of pounds. That was under the Act’s illegal content rules, not the children’s codes.

What happens next? The Secretary of State will decide whether to amend the law. Ofcom plans to publish an updated assessment of risks to children by January 2027. The report does not consider the government’s stated plan to ban social media for under-16s.

Questions the report leaves open

Which platforms have changed their feeds, and when will Ofcom measure the effects of those changes separately?

What accuracy standards will Ofcom set for identifying body stigma and depression content, and how will it address cases in which recovery-related content is mistakenly flagged?

How much would the new requirements cost services of different sizes, particularly those that host video and audio?

Will Ofcom collect platform data on these two categories?

Also Read:

39% of UK children bypass age checks, mostly by lying about their age

[googlesitekit-end-of-content]

WP QUADS Content Ad Plugin v. 3.0.5.1

For You

Read Reasoned by Nikhil Pahwa: How AI is changing our world

Sponsor a MediaNama Event