Client-Side Scanning Explained: the EU Chat Control Shift and How It Affects You

Imagine your phone reading every message, photo, and file you share before you even hit send.

That’s the core idea behind client-side scanning. It isn’t law in Europe yet. But on 9 July 2026, the European Parliament came close enough to make the direction unmistakable: a majority of MEPs voted against reviving warrantless message scanning, and it passed anyway. The permanent regulation that could mandate actual on-device scanning returns to the table in September 2026.

Some tech companies don’t even need to be told. They’ve been scanning voluntarily for years.

Here’s what this article covers:

• What client-side scanning (CSS) is and how it works technically
• Which EU and UK laws are driving it forward
• What the 9 July 2026 vote actually changed, and what it didn’t
• Which companies already scan your messages
• How you can protect yourself

Let’s dig in.

What is client-side scanning?

Client-side scanning (CSS) means checking the content of your messages directly on your device, before encryption kicks in.

Your phone or computer analyses text, images, or files and compares them against a database of flagged content. If something matches, it gets reported.

End-to-end encryption is sealing a letter in an envelope that only your recipient can open. Client-side scanning reads that letter while you’re still writing it. The envelope stays sealed, but your privacy is already gone.

Supporters say it’s needed to detect child sexual abuse material (CSAM). But privacy researchers argue it creates a mass surveillance system that affects everyone, including you.

One distinction matters before we go further. Most scanning happening in Europe today is server-side: it inspects unencrypted messages after they reach a provider. Client-side scanning is different, and more invasive, because it moves the inspection onto your own hardware. Keep that split in mind, because Europe’s two “Chat Control” files land on opposite sides of it.

How does client-side scanning work?

There are two main technical approaches. Both happen on your device, before your message is encrypted and sent.

Perceptual hash matching

Your device converts an image into a digital fingerprint (called a “hash”) and compares it against a database of known illegal images. If the hash matches, the content gets flagged and reported automatically. No human sees it unless the system triggers an alert.

Unlike a normal file checksum, a perceptual hash survives resizing, cropping, and recompression. That’s the point: it catches the same image even after it’s been altered. It’s also why the technique generalises so easily to content that isn’t CSAM at all.

Three systems dominate.

PhotoDNA (Microsoft)

The most widely deployed tool of its kind. It converts images into a greyscale grid-based signature and matches them against hash databases held by organisations such as NCMEC. Microsoft licenses it widely, which is why it shows up across platforms that didn’t build their own.

CSAI Match (Google)

Google’s system, built for video rather than still images. It fingerprints video content and matches it against known material, then routes hits to human review queues.

PDQ (Meta)

Meta’s own perceptual hashing algorithm, released as open source alongside a video counterpart. Because it’s public, it’s become a reference implementation for platforms building detection in-house.

AI classifiers

The second approach uses machine learning models to detect previously unknown illegal content by analysing visual and textual patterns.

The catch? AI classifiers produce high false positive rates. Innocent photos, such as medical images, family snapshots, or even art, can get wrongly flagged.

Hash matching asks “have we seen this exact thing before?” A classifier asks “does this look like the kind of thing we’re looking for?” The second question is far more useful to whoever controls the model, and far more dangerous to you.

Why neither can be limited to one purpose

Both methods share a deeper problem. Once scanning infrastructure sits on your device, nothing technically stops it from being expanded.

It could be repurposed to flag political speech, religious content, or anything a government decides is “objectionable.”

The Internet Society warns that CSS reduces security and privacy for law-abiding users while risking failure at the law enforcement goal it’s meant to serve. The EFF has shown why it’s technically impossible to build a scanning system that can only ever be used for one purpose. Once the mechanism exists, its scope becomes a policy decision, not an engineering one.

EU Chat Control: the legislation pushing for on-device scanning

The EU’s push for message scanning, dubbed “chat control” by critics, is the most significant threat to encrypted communications on the continent. But the story only makes sense if you keep two separate laws apart.

Chat Control 1.0: the voluntary derogation

In 2021, the EU introduced a temporary exception to its ePrivacy rules. It allowed tech platforms to voluntarily scan unencrypted messages for CSAM.

Services such as Gmail, Facebook Messenger, Skype, Snapchat, and Xbox messaging opted in. The exception was extended in 2024, moving its expiry to 3 April 2026.

Note what this regime is: server-side scanning of unencrypted content, at the provider’s discretion. It is not client-side scanning. “Voluntary” is also doing heavy lifting here. The choice belongs to the provider. Not to you, and not to any court.

Chat Control 2.0: the CSAR proposal

In May 2022, the European Commission proposed a permanent regulation called the Child Sexual Abuse Regulation. Critics quickly labelled it Chat Control 2.0.

This proposal goes much further. It would oblige providers to scan on the basis of administrative “detection orders” – and under the original draft, that reaches end-to-end encrypted services through client-side scanning.

This is the file that matters for CSS. The fifth trilogue ended on 29 June 2026 without agreement. Talks resume in September 2026, and the core dispute hasn’t moved: blanket scanning at the industry’s discretion, or targeted detection ordered by a judge against actual suspects.

Where things stand now

On 26 March 2026, the European Parliament voted 311 to 228 against extending the voluntary regime. The legal framework expired on 3 April 2026.

Then the procedure took over. On 7 July, Parliament fast-tracked an urgent procedure by 331 to 304. The vote itself landed on 9 July, the last voting day before summer recess, with roughly 112 MEPs already absent.

The result was a paradox. 314 MEPs voted to reject the extension. 276 voted to keep it, with 17 abstentions. More MEPs said no than yes. But a second-reading rejection requires an absolute majority of the full chamber: 361 votes. The motion fell 47 short, so the Council’s text stood by default.

Warrantless scanning of unencrypted messages is legal again until April 2028, or until a permanent regulation replaces it.

Two amendments tell you where Parliament’s centre of gravity actually sits. One would have restricted scanning to suspects identified by the judiciary: it won 322 to 255, and still died on the same threshold. The other passed, and expressly excludes end-to-end encrypted communications from the scope of Chat Control 1.0.

So WhatsApp, Signal and Telegram are now formally out of reach of the voluntary regime. That’s a real signal for encryption, and it’s worth being precise about its limits: it applies only to this track. CSAR is still the venue where client-side scanning obligations could reach encrypted services. A tactical win in one file, not a resolution.

One more wrinkle worth knowing: providers that kept scanning between April and July did so with no EU legal basis at all, and July’s vote doesn’t retroactively cure that gap.

Despite the legal uncertainty, Google, Meta, Microsoft, and Snap had already said they would continue scanning private messages.

The permanent CSAR regulation is still being negotiated, with talks expected to resume in September 2026. This is where governments across Europe and the UK continue to push for mandatory on-device scanning that could reach encrypted communications as well.

The UK Online Safety Act and on-device scanning

The EU isn’t alone. The UK’s Online Safety Act 2023 takes a parallel approach, and in some ways goes further.

What Ofcom can require

The Act gives Ofcom, the UK communications regulator, power to require platforms to deploy “accredited technology” for content detection. Privacy advocates warn this could mean on-device scanning of phones and other devices before messages are encrypted.

In December 2025, the government designated cyberflashing and content encouraging self-harm as priority offences under the Act, with the regulations taking effect in January 2026. This triggered stricter content detection obligations.

The authority to compel scanning is already on the statute book; what’s missing is the operational layer. Until Ofcom publishes its accreditation framework, those powers are dormant but not gone.

Note the contrast: while the EU has just carved encryption out of its voluntary regime, the UK has written powers that reach through it.

How Signal, WhatsApp and Apple responded

Signal and WhatsApp have both stated they would leave the UK rather than implement on-device scanning.

Apple withdrew Advanced Data Protection for UK users on 21 February 2025 after the government, using powers under the Investigatory Powers Act, demanded backdoor access to encrypted iCloud data. Rather than weaken its encryption, Apple disabled the feature in the UK.

The fundamental tension is clear. You can’t scan encrypted messages without either breaking the encryption or scanning content beforehand, and both options destroy the security model.

Which companies already scan your messages?

Several major tech companies already scan your interpersonal communications, not just social media or chat apps. The scanning covers email, cloud chat, gaming chat and direct messages.

Here’s who does it, what they use, and what it means for you.

Google

Scans: Gmail, Google Chat Uses: PhotoDNA and its own CSAI Match system

If you use Gmail for personal correspondence, your unencrypted messages sit inside the scope of the regime revived on 9 July. Google has said it will keep scanning.

Meta

Scans: Facebook Messenger, Instagram DMs Uses: PhotoDNA and its proprietary PDQ system

Meta scans the platforms where its encryption is optional or absent. Instagram DMs and Messenger threads without end-to-end encryption enabled are readable, and therefore scannable.

Microsoft

Scans: Outlook, Skype, Xbox messages Uses: PhotoDNA, which it developed

Microsoft built the most widely deployed hash-matching tool in the industry and applies it across its own consumer services, including Outlook email.

Snap

Scans: Snapchat messages Uses: hash-matching tools

Snapchat’s ephemeral design doesn’t exempt it. Messages are inspected before they disappear.

Scanned vs not scanned: where your messages stand

The exclusion for encrypted services applies to the voluntary regime only. The permanent CSAR proposal could still impose client-side scanning obligations on encrypted platforms.

*There is ongoing debate and some allegations about private‑message scanning on WhatsApp. WhatsApp publicly maintains that it cannot read end‑to‑end encrypted messages and opposes laws that would require client‑side or server‑side scanning of all chats; competitors and some commentators claim otherwise.

How to exit client-side scanning

Most people assume message scanning only happens inside chat apps or social media. In reality, it already reaches email, cloud chat, gaming chat and any service where people exchange content. Leaving this ecosystem means choosing services that don’t inspect your interpersonal communication.

1. Choose services that don’t rely on content scanning

Some platforms analyse your communication because their business model depends on inspecting user content. Others simply don’t.

Services that implement real end-to-end encryption cannot scan your messages, because they have no access to them. This is a technical property, not a marketing promise. Mailfence applies this principle to email, which is why your messages are not inspected.

Mailfence doesn’t scan your encrypted messages, because it can’t. True end-to-end encryption means even the service operator has no access to your content. That isn’t a policy we could quietly change next quarter. It’s a property of the architecture.

Our servers are in Belgium, under GDPR and Belgian jurisdiction. See our transparency report and warrant canary for what that means in practice.

2. Pick a provider whose business model doesn’t need your data

Large advertising-driven platforms scan messages because they need behavioural data. Privacy-focused services don’t.

Jurisdiction matters. A Belgian provider operates under GDPR and Belgian data‑protection law, which strictly limit how your message content can be repurposed. They can’t just decide to reuse it for broad scanning or profiling without a clear legal basis and compliance with purpose‑limitation and user‑rights rules. That’s a stronger protection than a terms‑of‑service promise alone.

3. Use true end-to-end encryption

Real end-to-end encryption only protects you when your device is free to encrypt without interference. Client-side scanning changes that model by inspecting content before encryption, which means encrypted or not, your message can still be read.

Mailfence avoids this entirely. Our OpenPGP implementation encrypts your email on your device.

Switching your email to a privacy-first service is one of the most impactful steps you can take. Your inbox contains years of personal, financial, and professional information. For more, see our email security guide and business email security best practices.

Final thoughts on client-side scanning

Once your device inspects what you type before encrypting it, the distinction between a private conversation and a monitored one stops being technical and becomes political.

The 9 July vote didn’t cross that line. It did something quieter and arguably worse: it showed that a measure previously rejected by a majority of MEPs can survive on a procedural threshold, during a recess, and buy itself two more years. The legislative file that actually reaches onto your device – the Client‑Side Scanning Regulation (CSAR) – returns to the table in September 2026.

If these rules come into force, there will be a clear “before” and “after”, and every service will have to decide how it responds. Some platforms may build client‑side scanning into their apps. Others won’t. Mailfence is in the second group, for structural rather than merely promissory reasons: when you use our end‑to‑end OpenPGP or password‑based encryption modes, your email content stays inaccessible to our infrastructure. Combined with Belgian jurisdiction and GDPR, which strictly limit how data can be repurposed, this makes content‑level scanning or profiling by Mailfence far harder to introduce than simply changing a terms‑of‑service.

Scanning now reaches email, chat, cloud storage, and gaming. Your choice of tools decides whether your messages stay private.