The UK has not announced one completed AI tool that detects child sexual abuse material (CSAM) across the internet. Instead, separate efforts are under way: funded research into detection in encrypted services, prospective AI-supported work to identify and categorise abusive imagery, and a legal framework for authorised testing of AI models. As of the government’s 3 September 2026 parliamentary answer, safeguards and operational arrangements for that testing regime were still being prepared.
What the UK government is proposing
The headline describes a policy-development effort, not the launch of a finished product. The initiatives cover different technical and legal tasks, and should not be treated as one system or as evidence that a new detection tool is already operating.
As an Amazon Associate I earn from qualifying purchases.
| Initiative | What it covers | Status established by official sources |
|---|---|---|
| Safety Tech Challenge Fund | Approaches to detecting or preventing abuse in encrypted communications, including known-content matching, on-device or real-time image and video detection, and age assurance. | Five project winners announced in November 2021 to explore these approaches; the announcement described project aims, not deployment of all five as products. UK government, November 2021 |
| International work announced by the Home Office | Tools to disrupt networks profiting from online child sexual abuse. Potential capabilities could include identifying AI-generated material and using AI to categorise imagery to reduce law-enforcement exposure. | Announced on 9 September 2025 as work with the US, Canada, Australia and New Zealand. The announcement described possible capabilities, not a deployed tool. Home Office, 9 September 2025 |
| Authorised AI model testing | Testing AI models for CSAM, non-consensual intimate imagery and extreme pornography by organisations authorised by the Home Secretary. | The Crime and Policing Act 2026 enables the regime. On 3 September 2026, the government said safeguards and operational arrangements were still being established and the regime would launch when safe and practical, subject to parliamentary scheduling. UK Parliament, 3 September 2026 |
| Border device scanning | Checking a device for material already known to the Child Abuse Image Database (CAID). | The Home Office says this capability had been operationalised at the border. It is distinct from identifying newly generated material or testing AI models. Home Office factsheet, updated 30 September 2026 |
Why “detection” can mean different things
Known-content matching and AI classification solve different problems. A system can compare a file with a reference database such as CAID, or attempt to classify material that is not already catalogued, including AI-generated imagery. Other approaches may look for abuse during live video or operate on a device before messages are encrypted. A model-testing regime, meanwhile, assesses AI models; it is not the same as scanning every user’s messages.
Free tools Windows power users keep installed
One-click scans. No signup required.
Those distinctions matter for privacy and effectiveness. Location and timing affect what a system can see: device-side checks, platform-side review, checks before encryption, and examination at a border are not interchangeable. Nor does a reported number of cases or intelligence leads establish a classifier’s accuracy. The UK sources do not provide a verified accuracy rate, false-positive rate or independent performance evaluation for the proposed authorised model-testing regime.
#1 Best Overall
What the reported figures do—and do not—show
- The Home Office factsheet reports Internet Watch Foundation figures of 3,443 AI-generated child sexual abuse videos identified in 2025, a 26,385% increase compared with 2024. It also reports that more than 15,000 AI-generated images and videos were assessed during 2024 and 2025. These figures indicate the scale of identified material; they are not performance measurements of the proposed UK testing regime. Home Office factsheet, 30 September 2026
- The same factsheet reports 311,610 reports confirmed as or linked to CSAM in 2025. That is a report count, not a count of unique images or victims. It says 15% of criminal imagery assessed was Category A. Home Office factsheet, 30 September 2026
- In border trials described by the factsheet, 44 consented device examinations produced actionable intelligence in 70% of examinations; 33% of that actionable intelligence was relevant to child protection. These are trial results from a small set of consented examinations, not detection accuracy, precision or recall figures for an AI model. Home Office factsheet, 30 September 2026
How the UK border phone scan differs
The Home Office says its border capability scans for material already known to CAID. A scan takes approximately 15 seconds, does not download the material, and alerts an officer if known material is present. This is database matching, not a claim that the scan can identify previously unseen or AI-generated imagery.
The Crime and Policing Act 2026 also allows officers to require a person entering or leaving the UK to unlock a device for examination when the officer reasonably suspects it may contain CSAM. That is a separate legal power from the technical description of the known-material scan. The factsheet does not establish that the scan identifies all illegal material on a device.
Rank #2
How the EU measure relates to the UK effort
The EU has a separate, temporary legal framework for voluntary provider detection and removal of CSAM. The Council of the EU said on 23 July 2026 that the measure had been reinstated through 3 April 2028, while longer-term rules remained under negotiation. This is EU context, not the legal basis for UK initiatives.
In the Council’s announcement, Ireland’s Minister for Justice, Home Affairs and Migration Jim O’Callaghan said: “Today is an important day for the protection of children. With the green light given, we have moved quickly to address the legal gap that existed when the derogation to the ePrivacy Directive lapsed in April. The decision taken by the Parliament earlier this month, and by the Council today, allows online providers to resume their voluntary efforts to fight the spread of child sexual abuse material online.” Council of the EU, 23 July 2026 The Council’s description of the interim measure and negotiations is available here.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What is still unknown
The public record described here does not establish when the authorised UK model-testing regime will begin, which organisations will be authorised, the final operating safeguards, or how its performance will be independently evaluated. It also does not establish that the prospective AI capabilities announced in 2025 have been deployed. Until those details are published, claims that a single new AI system is scanning encrypted services or reliably identifying newly generated CSAM go beyond what the announcements say.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
