This week I wrote about “mass surveillance in the name of protection”.
This is the Saturday/Sunday follow-up. (c) (r) ™ :-). (cross out whichever day doesn’t apply — or both, if you’re reading this on Monday!)
What I’m writing here may sound absurd — but it really isn’t.
Take a second and consider how your government has actually operated over the past 50 years.
Recap
To process 500 billion messages every day, you’d need an insane amount of hardware — not just to store all that data, but to process it, index it, and somehow make sense of it.
This would cost hundreds of millions of dollars annually and still generate an insane number of false positives, no matter how good your AI is.
Exactly as I wrote in the article mentioned above.
AI and text
AI will play an insanely important role in processing that flow.
But what happens when you throw some proverbial sand into the machine?
The text above can easily be processed by an LLM. It relies on the simple use of spaces between words to determine where one word ends and another begins.
So what happens when someone removes the spaces?
YOUCANREADTHISSENTENCEWITHOUTTHESPACESPRESENT.
But if I feed that to an LLM, it has absolutely no clue what to make of it. The tokenizer goes haywire!
And what if I combine this with a simple ROT13 (rotate by 13) algorithm?
LBHPNAERNQGUVFFRAGRAPRJVGUBHGGURFCNPRFCERFRAG
An AI or LLM won’t be able to process this, no matter what.
But if I apply ROT13 the other way around, I can read it again.
This is just basic obfuscation — not even real encryption — but it would still wreak havoc on detection algorithms.
Would this yield a false positive? Would I get a fine?
AI defeated.
Would this bill mean we’re now legally required to use proper spacing at all times — and that obfuscation isn’t allowed anymore?
AI and pictures
AI can also be used to inspect pictures — to see what’s in them and whether the content is “acceptable” or not.
But again, AI is extremely dependent on patterns in the image. Even changing its color palette can make the system far less efficient.
Now imagine splitting an image into 50×50 pixel blocks and rearranging them in a different order.
Would AI still be able to process it correctly?
With a bit of software, anyone could simply reconstruct the original image anyway.
JPEG uses compression and adds artifacts, which makes detection even harder.
Performing image processing in GIMP or Photoshop will alter the image itself, often causing detection algorithms to go completely haywire.
AI defeated.
Would this bill mean we’re only allowed to upload approved TIFF files — to make sure the algorithm can “see” properly — and that image processing would be banned altogether?
Certificates
We all use certificates to log in to our bank or basically any other website.
These certificates ensure that the traffic between the server and our computer is safe.
But if we use the same mechanism for chatting with each other, the government can’t tap into that data stream — unless someone builds in a backdoor…
Encryption defeated?
Would this bill mean we can no longer rely on certificates, or worse — would it ban their use entirely?
VPN
And then there’s the use of VPNs — virtual private networks — again dependent on certificates.
A bad actor could simply hand a certificate to another bad actor so they can connect to the same private network and chat freely.
Would this bill mean that VPNs could be banned altogether?
Technology
It took me less than twenty minutes to come up with the examples above for defeating AI detection algorithms. With a bit more thought, I could probably add another ten or twenty to that list.
And yes — I’d be that asshole who deliberately throws sand into the detection machine.
Not because I have something to hide, but because I genuinely believe the government doesn’t need to know everything.
And I know there are plenty of others who feel the same way.
2030 – The new 1984
You can laugh at the image: citizens queuing at local offices to hand in their VPN certificates, TLS officially deprecated, and national guidelines dictating how many spaces to use between words.
Funny and unrealistic — until you remember how often bad policy ends up walking in that exact direction once money and panic start to mix.
And remember: just the hardware required to process all those messages will cost hundreds of millions of dollars.
You can bet your ass that no government would tolerate those systems not working properly.
Absurd rules, yet plausible.
Imagine the “Digital Safety Directive, 2030” in practice:
- All uploaded images must be TIFF. JPEG and WEBP are “lossy obfuscation” and therefore banned.
- Any image edit (color corrections, inverted palettes, micro-tiling) is classified as “content obfuscation.” Offenders risk administrative sanctions.
- Citizens must register their VPN certificates with the Ministry of Secure Communications. Unregistered endpoints will be blocked by the Government Router™.
- Text submissions must have standardized spacing. Automated systems only accept single spaces; multiple consecutive characters without whitespace are flagged for human review, and possible can result in fines.
You chuckle, but look at history: outlier technical quirks have become regulation before. Governments don’t always invent plausible technical solutions — sometimes they legislate the simplest outward facts of life (how you format text, what file formats you may use) because those are easier to regulate than addressing the root problem.
Satire moment (press release)
Press release — EU Digital Safety Authority, 2030
To improve detection fidelity, citizens must submit images in TIFF format only. Palette transforms and micro-tiling are hereby classified as “content obfuscation.” Unauthorized edits will be fined. To appeal, request a Pixel Integrity Certificate; processing time: 6–8 weeks, costs: 100 euro’s!
It’s funny until you realize that policy often behaves like satire in real life — not because policymakers hate nuance, but because regulation loves enforceable simplicity instead of looking at the regulation itself if it is workable.
Takeaway
The whole chat-control debate is not about whether child abuse is real or whether law enforcement should have tools — of course they should!
The real question is what kind of tools and how they are used.
Mass, indiscriminate scanning of private messages trades away privacy and creates an enormous, expensive bureaucracy that will never catch the truly cunning criminals — but will certainly catch the rest of us.
This article sounds absurd.
But it really isn’t.
Please say NO to this route.
THISISMEBRAINSAYINGBRAINOUTANDAIHASNOFUCKINGCLUEWHATIWROTEHEREANDIHAVENTEVENUSEDROT13ONTHISSENTENCE-BRAINOUT!