Know what’s on your site before a plaintiff’s firm does
A free privacy audit shows which cookies, pixels and trackers are running — and which ones commonly appear in privacy claims.
Ofcom has told the UK’s biggest platforms, its file-sharing services and a much wider group of risky sites that waiting for users to report child sexual abuse material (CSAM) is no longer enough. On November 16, 2026, the regulator published its decisions on a package of “additional safety measures” under the Online Safety Act. The headline changes are that certain services should actively look for abuse imagery nobody has seen before, that hash matching for known material spreads to more services, including pornography sites, and that people who share this content should be banned. For trust-and-safety, privacy and legal teams, the work now is mapping which measures apply to them before the changes take effect, which Ofcom expects in early 2027.
TL;DR
- Ofcom published its Additional Safety Measures statement on November 16, 2026, adding new CSAM measures to its Online Safety Act codes of practice, subject to parliamentary scrutiny and expected to apply from early 2027.
- Large services at medium risk, services with more than 700,000 monthly UK users at high risk, and high-risk file-storage and file-sharing services should use proactive technology to detect previously unknown image-based CSAM.
- The changes sit alongside a separate duty, in force since April 7, 2026, for user-to-user services to report detected CSEA content to the National Crime Agency.

What Ofcom actually published
The announcement closes out consultations that Ofcom ran in April and June 2025. Its news release on the new CSAM rules lists three core steps: banning users who generate, upload or share child sexual exploitation and abuse (CSEA) content; using hash matching to find and remove known CSAM on more sites and apps, pornography services among them; and using automated tools to find previously unknown CSAM on certain platforms, including all file-sharing services and the largest social media services.
Oliver Griffiths, Ofcom’s Online Safety Group Director, said the regulator is “not satisfied that enough is being done.” He added that the rules “impose important additional legal duties on companies. And we will keep pushing platforms to do more to prevent, detect and remove this abhorrent material, report perpetrators, and shut them out for good.”
The statement goes beyond CSAM. Services should also keep illegal content out of children’s recommender feeds and strengthen anti-grooming defaults. On grooming, they can either protect all users or use highly effective age assurance to work out which accounts belong to children. Ofcom also expects hash matching for terrorist content, sanctions for UK users who share illegal content, and a way for users to report livestreams showing an imminent risk of physical harm. It dropped the children’s livestreaming rules it had consulted on, because the coming under-16 social media ban will keep many children off those services. Next spring it plans to consult on requiring proactive technology for a broader range of harms.
“Legal duties” or recommended measures?
Some early summaries described these as brand-new legal obligations, so precision matters here. The statutory duties already exist in sections 10 and 12 of the Online Safety Act 2023. What Ofcom has done is add recommended measures to its codes of practice. The government’s explanatory memorandum laid before Parliament says plainly that the instruments “do not create new statutory duties.”
The practical effect is still significant. A provider that follows every relevant code measure is treated as complying with the matching duty. A provider that takes another route can do so, but it has to keep a record of its alternative measures and explain how they meet the duty and protect users’ privacy and freedom of expression. In an investigation, Ofcom and the courts must take the codes into account. In practice, the codes set the baseline you will be measured against, which is why Ofcom calls them duties.
Who has to do what
Scope depends on two things in each service’s illegal content risk assessment: the risk level it has assigned for image-based CSAM, and its UK user numbers. The new proactive technology code, published as a standalone draft, applies to:
- large user-to-user services (more than 7 million monthly UK users) assessed at medium risk of image-based CSAM;
- user-to-user services with more than 700,000 monthly active UK users assessed at high risk; and
- file-storage and file-sharing services assessed at high risk, regardless of size.
The code doesn’t name a vendor or a single tool. It lets providers use text matching, application-specific machine learning, or a combination, “where technically feasible.” Whatever they choose has to meet a set of criteria: high-quality training data, steps to address bias, performance evaluation, safeguards against misuse, testing in context, ongoing maintenance, human review of an appropriate share of outputs and of undetected content, and a way to feed results back into the system. Hash matching already relies on one-way digital fingerprints of images that have been confirmed as abuse material. The Internet Watch Foundation says its list alone holds more than 3.2 million of them. Proactive detection is harder, because by definition it has to flag images that nobody has classified yet.
A separate requirement already applies. Since April 7, 2026, user-to-user services in scope have had to report detected CSEA content that hasn’t already been reported to the National Crime Agency. Content already sent to NCMEC’s CyberTipline counts as reported. Ofcom’s guide to the CSEA reporting duty notes that the duty applies whatever a service’s size or risk level, and that breaches can lead to penalties of up to 10% of qualifying worldwide revenue or £18 million, whichever is greater. More detection means more reports, so the two regimes have to be planned together.
The privacy tension nobody can skip
Scanning every upload for unknown abuse material means processing personal data about every uploader, not only offenders. The code itself acknowledges that implementation “will inevitably involve the processing of personal data.” It points to the Information Commissioner’s Office, which expects a data protection by design and by default approach. Several UK GDPR questions follow:
- Lawful basis. Code measures are recommendations, not direct legal commands, so teams should document whether they rely on legal obligation, legitimate interests or both, and why.
- Criminal offence data. A detection, a ban or an NCA report links a person to suspected criminal activity. That triggers Article 10 UK GDPR and the conditions in the Data Protection Act 2018.
- Automated decisions. Banning an account on the strength of a classifier alone can be a solely automated decision with significant effects. The human review built into the code helps, but only if it is real.
- Accuracy and false positives. A wrongly flagged family photo can end with a ban and a police report. The updated codes also expect providers to correct mistakes where they wrongly identify content as illegal.
- Private messaging. Schedule 4 of the Act bars Ofcom’s codes from recommending proactive technology that analyses user-generated content communicated privately, or its metadata. The debate over end-to-end encrypted messaging isn’t settled by this statement.
The same conflict is playing out elsewhere. In the United States, West Virginia’s lawsuit against Apple over iCloud and CSAM tests whether a company can be blamed for not scanning. In Brussels, the long-running fight over the EU’s Chat Control proposal turns on how far detection orders may reach. The UK approach is narrower than both. It ties obligations to risk assessments and user numbers instead of applying blanket scanning.
A compliance checklist for in-scope services
Teams that want to be ready before the measures take effect can work through these steps in order:
- Re-run the CSAM risk assessment. Your image-based CSAM risk rating now decides whether proactive detection applies, so check the evidence behind it.
- Count UK users properly. The 700,000 and 7 million thresholds use monthly active UK users, calculated the way the code describes, not marketing figures.
- Decide on hash matching coverage. If you run a pornography service or another newly covered service, plan which hash lists you will use, how often they update, and how much human review you will apply.
- Choose and test detection technology. Write down why you chose text matching, machine learning or both, and keep the bias, performance and contextual test results.
- Write the ban policy. Define what triggers a ban for generating, uploading or sharing CSEA, how you stop banned users from returning, and how appeals work.
- Connect detection to reporting. Route every detection into an NCA or NCMEC reporting workflow, without duplicate reports.
- Complete a DPIA. Large-scale, systematic monitoring of user content is a textbook case. Record the lawful basis, Article 10 conditions, retention periods and safeguards against false positives.
- Update your records. Add the new processing to your records of processing activities and your privacy information, including how long hashes, flagged items and reports are kept.
Why AI-generated material raises the stakes
Ofcom’s release cites Internet Watch Foundation research reporting a 26,385% annual increase in AI-generated child sexual abuse videos online. Hash lists catch only material that has already been found and classified, while generative tools can produce new images endlessly. That’s the main reason the code now targets “unknown” content. It also explains why privacy regulators have warned about synthetic imagery of real people: the same models that create new abuse material make detection a moving target.
Child protection groups welcomed the package. The NSPCC said police data shows an average of 100 child sexual abuse image crimes a day across the UK. The Internet Watch Foundation said “the net is tightening.” Industry responses to the original consultation raised concerns about prescriptive requirements, cost, technical feasibility and privacy, according to the government memorandum.
Frequently asked questions
When do the new measures apply?
Ofcom submitted the draft code amendments to the government on November 16, 2026, and they have been laid before Parliament. Ofcom expects them to come into force from early 2027, after the parliamentary process is complete.
Does this apply to services based outside the UK?
Yes. The proactive technology code applies whether or not the provider is in the UK, provided the service falls within the Online Safety Act’s scope.
Is a small file-sharing service covered?
If it has assessed itself as high risk for image-based CSAM, yes. That category has no user-number threshold.
Can we use a different method instead of the code’s measures?
You can, but you lose the automatic presumption of compliance. You must document your alternative and show it meets the duty while protecting privacy and freedom of expression.
Do we report to the NCA if we already report to NCMEC?
Not for the same content. Ofcom says content already reported to NCMEC is not “unreported,” and NCMEC passes UK-linked reports to the NCA.
The net only works if you can explain the knots
Ofcom has made clear it wants detection, removal, bans and reporting to work as one system. The services that do best won’t be the ones that buy the most powerful classifier. They will be the ones that can show the regulator and the ICO how each knot in that net was tied: why the risk rating is what it is, why the tool fits, who reviews its decisions, and what happens to the data afterwards.
Put this into practice
Turn privacy guidance into working controls.
Captain Compliance helps teams discover tracking technologies, enforce consent, manage privacy requests and document the evidence behind every decision.