You might have seen some fairly astonishing headlines recently, announcing that the UK’s speech regulator, Ofcom, has decided not to force companies to scan people’s phones, and that the new “safety measures” announced on November 16 are actually good news.
What Ofcom has done is update two of its Codes of Practice – the Illegal Content Codes and the Protection of Children Codes. The changes are detailed in a document published on the regulator’s website, and are set to come into force in February 2027, if Parliament approves them.
Ofcom’s “recommendations” – non-compliance with which can result in fines of up to 10 percent of a company’s annual global turnover – include using “proactive technology” to detect illegal content, and prevent children from accessing it.
They also include permanent bans for users found to have shared content that shows child abuse – and using “highly effective” age assurance to decide which accounts should have anti-grooming protections.
In the case of encrypted group chats that Ofcom wants companies to police, the regulator is aware that this "may interfere significantly with users' privacy, in particular where lawful private communications are removed or where an incorrect illegal content judgement results in inaccurate inferences about users or disclosure of information about a user to law enforcement."
The Record writes that the UK’s communications regulator Ofcom “does not have the power to make technology companies scan people’s personal devices for child abuse images.”
What Ofcom actually said was that it “does not have powers under the Act to regulate devices.”
While the latest Ofcom “recommendations” – or “safety measures” – don’t include client-side scanning, the regulator can still order that once it’s “accredited” a tool that does the scanning.
Section 121 of the Act allows Ofcom to give a “technology notice” to a company, ordering it to use “accredited” software to find CSAM in private messages.
Our reading of the Act is that Ofcom doesn’t have to regulate a device – it regulates the app. And the app is on the device. And once the app is ordered to scan for illegal content in private messages, where else can it do it but on the device? In this case, the phone.
The government's accuracy standards have been in force since September, and Ofcom has yet to say that any tool has been accredited.
The regulator has been working on this for a while, and in 2024, we reported that Ofcom said it had the power to “make a provider use (or in some cases develop) a specific technology to tackle child sexual abuse or terrorism content on their sites and apps.”
The problem with that is that even without scanning a single message, these rules already reach into those encrypted chats. The Record writes that for those, Ofcom's revised "guidance" allows the platforms to judge a whole group based on its profile picture, description, or coded terms - as long as that judgment is "backed by a British police report."
But what about the security exemption from the scanning regime that Ofcom is putting in place, one "recommendation" at a time? Ofcom itself says that the rules don't apply if "it is not technically feasible to analyse user-generated content." "This is particularly the case where such changes would materially compromise the security of that service."
But don't get your hopes up. Ofcom immediately qualifies this by saying that it won't accept that implementing the scanning measures may be infeasible "because doing so may require a change in the design and/or operation of a service." And neither will it accept that "high costs" are a valid reason.
And who decides what is or isn't technically feasible? Ofcom, of course: "Where a service provider claims that it is technically infeasible for it to implement proactive technology, we have the power to investigate this."
One of the measures that have been added to the already existing ones is the scanning of public posts for images that are brand new and have never been seen before - to check if they are illegal abuse images.
The code, which is one of the "recommended" measures that companies will ignore at their own peril, concerns large user-to-user services (over 7 million monthly active UK users) that are considered at medium risk of image-based CSAM. It also covers user-to-user services with more than 700,000 monthly active UK users that are at high risk of the same, as well as high-risk file-storage and file-sharing services.
In fact, the regulator is so aware of the fallibility of the tools it is effectively imposing on tech companies that it has included in the code "steps providers should take to reverse incorrect content decisions" - a measure that is said to have been added "in response to concerns raised by stakeholders around the risk of false positives."
Ofcom explains that terrorism hash matching will be required for services where users can post photos, videos and other visual material, and are at medium or high risk of hosting terrorism content. This applies to large services, and those with more than 700,000 monthly active users in the UK who are at high risk.
Ofcom's own 2022 technical report on perceptual hashing said that perceptual hashes have "a non-negligible risk of false positives and false negatives, i.e. dissimilar input files may have identical or similar hashes while similar input files may have dissimilar hashes."
Ofcom concedes, "terrorism offences are often contextual; there might be legitimate reasons for sharing content related to terrorism, such as journalism."
The Open Rights Group told Ofcom during the consultation about "the lack of published error rates for terrorism hash matching."
As for the lists of content that companies will be matching against - Ofcom first said that databases compiled by outside entities were not suitable because they "were not limited to hashes of terrorism content as defined in the Act."
But then it changed its mind and went for Tech Against Terrorism’s Hash List API, saying that content that is not a UK terrorism offence “is either capable of being filtered out by services or presents a low risk of comprising more than a small proportion of the database.”
Ofcom also mentions the Global Internet Forum to Counter Terrorism (GIFCT), but notes that both extend beyond UK "proscribed organisations."
GIFCT's 2024 report said that the HSDB contains "approximately 408,000 unique and distinct items." Of the hashes with behavioral labels, "the majority (75.94%) focused on glorification of terrorist acts."
The Brennan Center said in its own report that "we know almost nothing about what's actually in the database, how often content is incorrectly flagged, or how often users are filing appeals."
Ofcom also admits that its new rules will result in legal posts being hidden from teenagers’ feeds. Ofcom’s own words are: “there is a likelihood that a large volume of legal content will be excluded. This may significantly interfere with users’ rights, including highly protected forms of expression such as political speech.”
The anti-grooming protections cover services that are at high risk of “grooming” and large services, with 7 million or more monthly users in the UK, that are at medium risk. Ofcom wants these to either use “highly effective age assurance” or apply the measures to all users.
Ofcom also wants platforms to ban users who share CSEA content. “For adult users, the default expectation is a permanent ban, except in extremely limited and exceptional circumstances,” it says.
We don’t have to speculate about how this works – we already know. In May 2026, a Reform UK campaign video by Zia Yusuf was deleted by TikTok after a report through the Online Safety Act’s reporting mechanism. The notice he received read, “We have detected this policy violation based on a report that the content violated our Community Guidelines.” TikTok also warned him that any further violations may result in a permanent ban.
On October 20, 2026, Ofcom chief Melanie Dawes told MPs that she wants rules that “require proactive technology to detect illegal content more than we’re able to do at the moment.”
This is all in line with the “additional safety measures” that Ofcom announced on November 16, 2026 – and next Spring, Ofcom said, it will “consult on proposals that service providers should use proactive technology capable of detecting a broader range of harms.”
On client-side scanning, Ofcom said it “will keep its measures under review in light of any future legislative changes” – which means that the door is not shut, it’s just temporarily closed until a suitable tool is “accredited.”
This is a regulator that is never finished, and never satisfied. The “safety” machine it is building one recommendation at a time is already massive, and is getting bigger.










