Global content moderation

Ofcom's CSAM Hash-Matching Expansion Is Defensible, but Its Push Toward Mandated Tools for Unknown Content Needs Hard Safeguards

Ofcom's 16 November 2026 package extends hash-matching and user bans, a proportionate step. Mandating classifiers for unknown content is a different and riskier bet.

Ofcom's CSAM Package in Numbers People of Internet Research · Global 26,385% AI abuse video annual rise IWF research cited by Ofcom in its… ~100 Child abuse image crimes daily NSPCC average across the UK, quote… 290,000+ CSAM web pages found 2024 The most the IWF has recorded in i… peopleofinternet.com
Ofcom's CSAM Package in Numbers People of Internet Research · Global 26,385% AI abuse video annual rise ~100 Child abuse image crimes daily 290,000+ CSAM web pages found 2024 peopleofinternet.com

Key Takeaways

On 16 November 2026, Ofcom announced a package of new Online Safety Act measures. Services would have to ban users who generate, upload or share child sexual exploitation and abuse content, use hash-matching to find and remove known child sexual abuse material (CSAM) on more sites and apps, including pornography services, and apply the same method to known terrorist content. Some services would also have to use automated tools to find previously unknown CSAM. Ofcom expects the duties to take effect from early 2027, subject to parliamentary process.

The package bundles measures with very different risk profiles. The first two are defensible. The third needs far more scrutiny.

The strongest case for Ofcom

The case for acting is serious. Ofcom's announcement cites Internet Watch Foundation research finding a 26,385% annual increase in AI-generated child sexual abuse videos online, and the NSPCC's Rani Govender is quoted as saying there are an average of 100 child sexual abuse image crimes across the UK every single day. The IWF has separately reported that it found over 290,000 web pages containing CSAM in 2024, the most it had recorded in its 29-year history, according to its own account. Generative AI is making volume and novelty worse at once. A regulator that did nothing in that environment would be failing.

There is also a fair argument that voluntary adoption has been uneven. If one large file-sharing service runs hash-matching and a competitor does not, abusers simply migrate. A uniform floor removes that arbitrage.

Why hash-matching is the proportionate core

Hash-matching compares a fingerprint of an uploaded file against a list of material that human analysts have already assessed as illegal. Its design is what makes it easier to defend than most moderation technology. It looks for known items rather than judging context, so it does not ask a model to guess whether a photo of a child in a bath is abuse or a family snapshot. The IWF's Hannah Swirsky calls it, in Ofcom's release, "a tried and trusted form of child sexual abuse material detection." Its false-positive profile is narrow and its lists are curated by an accountable body.

The obligation is also not new in kind. Ofcom's illegal harms statement, first published on 16 December 2024 and updated on 10 November 2026, already sits behind the codes, and the IWF notes that all file-storage and file-sharing services were expected to do hash-matching regardless of size. Extending the technique to more sites, including pornography services, is an incremental widening of an established practice. Banning accounts that knowingly share CSAM is similarly hard to object to: the Act's illegal-content duties exist to remove precisely this conduct.

Extending the same hash-matching to terrorist content deserves more caution, and the reason is structural. CSAM is illegal in every context, so a hash list has a single clear referent. Terrorist content is not like that. The same footage can be propaganda in one upload and journalism, research or documentation of war crimes in another. A hash cannot tell the difference. If Ofcom proceeds, it should require that matched terrorist content be reviewed by a human in contexts where news organisations, researchers and archives may be affected, and that list governance be published.

Where the proportionality argument weakens

The unknown-content requirement is the real issue. Ofcom says automated tools should proactively identify and remove previously unknown CSAM on certain platforms, including all file-sharing services and the biggest social media. It also says it will consult by spring next year on directing platforms to adopt specific technologies rather than letting them self-assess.

That shift matters. Today, the Act's model is risk assessment followed by proportionate measures. A regulator that names the technology it expects skips that step, and it takes accountability for the technology's error rate. Classifiers that detect novel abuse imagery are far less precise than hash lists. They will flag lawful images, and they will do so at the scale of billions of uploads. The Electronic Frontier Foundation put the general problem bluntly on 27 October 2026: "perfect (balanced, targeted, transparent, swift and swiftly appealable) content moderation at scale is impossible." That post is about elections, not CSAM, but the arithmetic is the same. A tiny error rate applied to a very large volume produces many wrongly removed items and wrongly banned users.

The interaction with the ban requirement raises the stakes. Removing a file is recoverable. Banning an account for sharing CSAM, on the strength of a classifier's guess about unknown content, is a serious sanction against a person who may have done nothing wrong.

What proportionate implementation looks like

Ofcom's consultation should therefore be treated as the real decision point, and it should insist on a few things:

The breadth of Ofcom's powers is also worth keeping in view. The Act allows fines of up to 10% of turnover or £18 million, whichever is higher, so small services will comply aggressively and err toward over-removal rather than risk it.

Bottom line

Ofcom is right to widen hash-matching and to back bans for those who share abuse material. These are narrow, tested tools aimed at unambiguous illegality. The harder questions are the move toward specifying technology for unknown content and the extension of hash methods to terrorist material. Both can be defended only with human review, transparency and appeals built in from the start. Parliament has until early 2027 to insist on them.

Sources & Citations

  1. Ofcom: Platforms must take stronger action to combat CSAM (16 Nov 2026)
  2. Ofcom: Statement on protecting people from illegal harms online
  3. IWF: Online Safety Act illegal harms duties
  4. EFF: Social Media When Elections Are Close