[fwn]

> CSAM is also a list of hashes for some of the worst CP video/images out there. It doesn't read anything, just hash matching.

The list presumably contains CSAM hashes. However, it could also include hashes for other types of content.

AFAIK the specific scope at any point in time is not something that can be fully evaluated by independent third parties, and there is no obvious reason why this list could not be extended to cover different types of content in the future.

Once it is in place, why not search for documents that are known to facilitate terrorism? What about human trafficking? Drug trafficking? Antisemitic memes spring to mind. Or maybe memes critical of some government, a war, etc.

This is because, despite the CSAM framing, it is essentially a censorship/surveillance infrastructure. One that is neutral with regard to content.

[EagnaIonat]

CSAM scanning has been around for at least 15 years. All service providers are required to do it by law.

You are absolutely correct with your "what-ifs" and this underlines the need for more oversight and transparency.

The process (my knowledge is a few years old) is that service providers or Law enforcement from countries can submit files to the CSAM database.

The database is owned by National Center for Missing & Exploited Children (NCMEC).

Once they receive the files they review them and confirm that the files meet the standard for the database, document its entry, create a hash and add that to the database. After that the file is destroyed.

This whole process requires multiple approvals and numerous humans review the files before the hash goes into the CSAM.

Also every hash has a chain of custody. So in the event of an investigation they know exactly everyone who was involved in putting that hash into CSAM.

So it's possible to submit an image that is not what CSAM is intended for, but the chances of it even remotely getting into the database is next to nothing. To add to this service providers can be sued for submitting invalid files.

[fwn]

> CSAM scanning has been around for at least 15 years. All service providers are required to do it by law.

That is true for scanning in the cloud, but it's important not to conflate this with client-side scanning. The distinction between cloud and local processing is foundational. Collapsing that boundary would mark a serious shift in how surveillance infrastructure operates.

> Once they receive the files they review them and confirm that the files meet the standard for the database, document its entry, create a hash and add that to the database. After that the file is destroyed.

That is already a structural problem: If the original is destroyed, how can independent parties verify that database entries still correspond to the intended legal and ethical scope? This makes meaningful oversight functionally impossible.

Even if centralizing control in a state-funded NGO were considered acceptable (which is already questionable), locating that NGO in the US (subject to US law and politics!) is a serious issue. Why should, say, the local devices of German citizens be scanned against a hash list maintained under US jurisdiction?

> So it's possible to submit an image that is not what CSAM is intended for, but the chances of it even remotely getting into the database is next to nothing. To add to this service providers can be sued for submitting invalid files.

Procedural safeguards are good, but they don't solve the underlying problem: the entire system hinges on policy decisions that can change. A single legislative change is all it takes to expand the list’s scope. The current process may seem narrow today, but it offers no guarantees about tomorrow.

We’ve seen this pattern countless times: surveillance powers are introduced under the pretext of targeting only the most heinous crimes, but once established, they’re gradually repurposed for a wide range of far less serious offenses. It is the default playbook.

[EagnaIonat]

> That is true for scanning in the cloud, but it's important not to conflate this with client-side scanning.

From what you say it's clear you never read Apples paper on this.

The client puts a flag on a match. It is only verified on the server both by another scan and a law enforcement.

If the client doesn't flag a file, it can never be decrypted on the server by anyone except the device owner.

The current system just checks everything. If your device never talks to the cloud in both scenarios nothing happens.

> That is already a structural problem:

You seem to have an over simplified view of how it all works. They don't just throw hashes in.

They can verify it by the chain of custody and documentation that is stored about that hash.

> the local devices of German citizens be scanned against a hash list maintained under US jurisdiction?

CSAM is a UN protocol that has 176 countries signed onto it. Including Germany.

Many countries also have their own independent department that works with CSAM. Germany has their Federal police (BKA) that work that role. They work with NCMEC on ensuring the CSAM hashes are correct. Germany is also one of the strictest countries in relation to CSAM.

> the entire system hinges on policy decisions that can change.

Again it's an over simplification. If the US government did do that.

- It would first be challenged in the courts.

- They would not be able to hide the fact they have changed it.

- This would lead to service providers not assisting with the corrupted CSAM.

- As this is a worldwide initiative the rest of the world can just disconnect the US from the CSAM until what is put in is confirmed.

> It is the default playbook.

If they wanted to do that, the CSAM database is the worst way to do it.

I'd recommend you read up on all of it a bit more. Most of your claims are unfounded in relation to the CSAM.

[fwn]

> From what you say it's clear you never read Apples paper on this. ... You seem to have an over simplified view of how it all works. They don't just throw hashes in. ... Again it's an over simplification. If the US government did do that. ... I'd recommend you read up on all of it a bit more. Most of your claims are unfounded in relation to the CSAM.

The posturing about supposed expertise adds nothing. If you want to make an argument, make it. Vague appeals to technical depth are just noise.

> The client puts a flag on a match. It is [...] verified on the server [...] The current system just checks everything.

Sure, that’s how the flagging process works. It’s also beside the point. Listing technical details doesn’t change the core issue: this system performs scanning on the user device, which is what makes it problematic.

> If the client doesn't flag a file, it can never be decrypted on the server by anyone except the device owner. [...] If your device never talks to the cloud in both scenarios nothing happens.

Correct, but not relevant here. No one is arguing that airgapped devices leak information. The issue is what happens when devices are online.

> [On the structural problem of inability of independent oversight] They can verify it by the chain of custody and documentation that is stored about that hash.

What specific documentation would allow actual evaluation? And who can access it? The process is opaque by design: The list of neural hashes is private, matching and flagging happen silently, and escalation logic like threshold levels or safety voucher generation is not open to inspection. Whatever theoretical accountability might exist, it’s irrelevant in a system of systematic secrecy that cannot be independently observed or audited.

> CSAM is a UN protocol [...] countries [...] work with NCMEC on ensuring the CSAM hashes are correct. Germany is also one of the strictest countries in relation to CSAM.

Yes, Germany has police and ofc works to fight CSAM. That doesn’t change the concern: the system design is extensible and unverifiable. If a U.S. administration wanted to expand the scope (say, for terrorism, extremism, drugs, or IP enforcement), who exactly stops them? Not a German agency. Certainly not NCMEC.

> [On the obvious loophole of policy change] If the US government did do that. - It would first be challenged in the courts.

That is... optimistic. What legal mechanism exactly would allow a challenge to a (as an example) classified National Security Letter expanding the hash set? What court has even the standing to hear that? What precedent makes you believe such a challenge would surface in time?

> - They would not be able to hide the fact they have changed it.

Why not? The hashes are not reversible. The list is not public. The matches are not auditable. Gag orders are legal. What in this system ensures visibility or accountability?

> - This would lead to service providers not assisting with the corrupted CSAM. - As this is a worldwide initiative the rest of the world can just disconnect the US from the CSAM until what is put in is confirmed.

The Apple proposal is not a worldwide initiative, but a US-driven proposal involving a handful of US orgs. EU involvement in the whole issue has been comparatively lacking and is often dependent on US lobbying and funding. The idea that the world could or would opt out assumes a degree of transparency and technical independence that simply does not exist on this planet right now.

If you want to argue that the system is technically robust against political misuse, then please do. If there are decent guardrails in place, I'd really truly do like to know about them. But so far, it mostly reads like a wish list.