this post was submitted on 29 Jun 2025
85 points (89.7% liked)

Privacy

39370 readers
743 users here now

A place to discuss privacy and freedom in the digital world.

Privacy has become a very important issue in modern society, with companies and governments constantly abusing their power, more and more people are waking up to the importance of digital privacy.

In this community everyone is welcome to post links and discuss topics related to privacy.

Some Rules

Related communities

much thanks to @gary_host_laptop for the logo design :)

founded 5 years ago
MODERATORS
k_o_t@lemmy.ml
tmpod@lemmy.pt
Yayannick@lemmy.ml
ranok@sopuli.xyz
85
AI to make us more private? (lemmy.zip)
submitted 1 day ago by dodgeflailimpose@lemmy.zip to c/privacy@lemmy.ml
49 comments fedilink hide all child comments
 

Just listened to Naomi Brockwell talk about how AI is basically the perfect surveillance tool now.

Her take is very interesting: what if we could actually use AI against that?

Like instead of trying to stay hidden (which honestly feels impossible these days), what if AI could generate tons of fake, realistic data about us? Flood the system with so much artificial nonsense that our real profiles basically disappear in the noise.

Imagine thousands of AI versions of me browsing random sites, faking interests, triggering ads, making fake patterns. Wouldn’t that mess with the profiling systems?

How could this be achieved?

top 49 comments
sorted by: hot top controversial new old
[–] upstroke4448@lemmy.dbzer0.com 7 points 19 hours ago* (last edited 19 hours ago)

This strategy of generating fake data just doesn't work well. It requires a ton of resources to generate fake data that can't be easily filtered which ends up making the strategy non viable on most situations. Look at Mullvads DAITA and how it constantly has to be improved to fight this and, that's just for basic protection.

There is a bit of a cognitive dissonance that goes on, where people seem to understand that you are tracked constantly online and offline through all sorts of complex means but still think relatively mundane solutions could break that system.

[–] moseschrute@lemmy.ml 98 points 1 day ago* (last edited 1 day ago) (3 children)

I feel like I woke up in the stupidest timeline where climate change is about to kill us, we decide stupidly to 10x our power needs by shoving LLMs down everyone’s throats, and the only solution to stay private is to 10x our personal LLM usage by generating tons of noise about us just to stay private. So now we’re 100x ing everyone’s power usage and we’re going to die even sooner.

I think your idea is interesting – I was also thinking that same thing awhile back – but how tf did we get here.

[–] Reverendender@sh.itjust.works 33 points 1 day ago (1 children)

but how tf did we get here

With capitalistic gusto! 🤮

[–] tisktisk@piefed.social 13 points 1 day ago

Just buy more? Idk I'm all out of ideas

[–] blargh513@sh.itjust.works 4 points 1 day ago

There are ais that can detect use of ai. This is a losing strategy as we burn resources playing cat and mouse.

As with all things greed is at the root of this problem. Until privacy has any legislative teeth, it will continue to be a notion for the few and an elusive one at that.

[–] octobob@lemmy.ml 3 points 1 day ago* (last edited 1 day ago) (1 children)

Yeah agreed. What's going on in my state of Pennsylvania is they're reopening the Three Mile Island nuclear plant out near Harrisburg for the sole reason of powering Microsoft's AI data centers. This will be Unit 1 which was closed in 2019. Unit 2 was the one that was permanently closed after the meltdown in 1979.

I'm all for nuclear power. I think it's our best option for an alternative energy source. But the only reason they're opening the plant again is because our grid can't keep up with AI. I believe the data centers is the only thing the nuke plant will power.

I've also seen the scale of things in my work in terms of power demands. I'm an industrial electrical technician, and part of our business is the control panels for cooling the server racks for Amazon data centers. They just keep buying more more and more of them, projected til at least 2035 right now. All these big tech companies are totally revamping everything for AI. Like before a typical rack section might have drawn let's say 1000 watts, now it's more like 10,000 watts. Again, just for AI.

[–] moseschrute@lemmy.ml 2 points 1 day ago

Totally agree nuclear is a great tool but totally being used for the wrong purpose here. Use those power plants to solve our existing energy crisis before you crate an even bigger energy crisis.

[–] rumba@lemmy.zip 7 points 21 hours ago* (last edited 2 hours ago) (1 children)

This is a dangerous proposition.

When the dictatorship comes after you, they're not concerned about the whole of every article that was written about you All they care about are the things they see as incriminating.

You could literally take a spell check dictionary list, pull three words out of the list at random and feed it into a ollama asking for a story with your name that included the three words as major points in the story.

Even on a relatively old video card, you could probably crap out three stories a minute. Have it write them in HTML and publish the site map into major search engines on a regular basis.

EDIT: OK this was too fun not to do it real quick!

~ cat generate.py

import random
import requests
import json
import time
from datetime import datetime

ollama_url = "http://127.1:11434/api/generate"
wordlist_file = "words.txt"

with open(wordlist_file, 'r') as file:
    words = [line.strip() for line in file if line.strip()]

selected_words = random.sample(words, 3)
theme = ", ".join(selected_words)

prompt = f"Write a short, imaginative story about a person named Rumba using these three theme words: {theme}. The first word is their super power, the second word is their kyptonite, the third word is the name of their adversary.  Return only the story as HTML content ready to be saved and viewed in a browser."

response = requests.post(
    ollama_url,
    headers={"Content-Type": "application/json"},
    data=json.dumps({"model": "llama3.2","prompt": prompt})
)

story_html = ""
for line in response.iter_lines(decode_unicode=True):
    if line.strip():
        try:
            chunk = json.loads(line)
            story_html += chunk.get("response", "")
        except json.JSONDecodeError as e:
            print(f"JSON decode error: {e}")



timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
filename = f"story_{timestamp}.html"

with open(filename, "w", encoding="utf-8") as file:
    file.write(story_html)

print(f"Story saved as {filename}")

~ cat story_20250630_130846.html

<!DOCTYPE html>
<html>
<head>
<title>Rumba's Urban Adventure</title>
<meta charset="UTF-8">
<style>
body {font-family: Arial, sans-serif;}
</style>
</head>
<body>

<h1>Rumba's Urban Adventure</h1>

<p>Rumba was a master of <b>slangs</b>, able to effortlessly weave in and out of conversations with ease. Her superpower allowed her to manipulate language itself, bending words to her will. With a flick of her wrist, she could turn a phrase into a spell.</p>

<p>But Rumba's greatest weakness was her love of <b>bungos</b>. The more she indulged in these sweet treats, the more her powers wavered. She would often find herself lost in thought, her mind clouded by the sugary rush of bungos. Her enemies knew this vulnerability all too well.</p>

<p>Enter <b>Carbarn</b>, a villainous mastermind with a personal vendetta against Rumba. Carbarn had spent years studying the art of linguistic manipulation, and he was determined to exploit Rumba's weakness for his own gain. With a wave of his hand, he summoned a cloud of bungos, sending Rumba stumbling.</p>

<p>But Rumba refused to give up. She focused her mind, channeling the power of slangs into a counterattack. The air was filled with words, swirling and eddying as she battled Carbarn's minions. In the end, it was just Rumba and Carbarn face-to-face.</p>

<p>The two enemies clashed in a spectacular display of linguistic fury. Words flew back and forth, each one landing with precision and deadliness. But Rumba had one final trick up her sleeve - a bungo-free zone.</p>

<p>With a burst of creative energy, Rumba created a bubble of pure slangs around herself, shielding her from Carbarn's attacks. The villain let out a defeated sigh as his plan was foiled once again. And Rumba walked away, victorious, with a bag of bungos stashed safely in her pocket.</p>

</body>
</html>

Interesting that it chose female rather than male or gender neutral. Not that I'm complaining, but I expected it to be biased :)

[–] Eyedust@lemmy.dbzer0.com 3 points 16 hours ago

Yup, you'd be surprised what you can accomplish with 10gb of VRAM and a 12b model. Hell, my profile pic (which isn't very good, tbf) was made on that 10gb VRAM card using localhosted stable diffusion. I hate big corp AI, but I absolutely love open market and open source local models. Gonna be a shame when they start to police them.

To OP: The problem is that they're looking for keywords. With the amount of people under surveillance these days, they don't give a rat's ass if you went to your favorite coffee roasting site, they want to find the stuff they don't want you to do.

Piracy? You're on a list. Any cleaning chemical that can be related to the construction of explosives? You're on a list. These lists will then tack on more keywords that pertain to that list. For example, the explosives list will then search for matching components bought within a close span of time that would indicate you're making them. Even searching for ways to enforce your privacy just makes them more interested.

So then you put out a bunch of fake data. This data happens to say you viewed a page pertaining that matching component. Whelp, that list just got hotter and now there are even more eyes on you and they're being slightly more attentive this time. Its a bad idea. The only way you're getting out of surveillance, at least online, is to never go online.

In reality, they probably won't even do anything about the above. What they really want is money. Money for your info; money to sell more things to you. They want the average home to be filled with advertisements tailored from your information. Because those adverts make those companies money, which they then use to buy more information to monetize your existence. Its the largest pyramid scheme known to humanity, and we're the unpaid grunts.

The moment the world became connected through telephones, cable TV, and then internet this scheme was already in motion way beforehand. Let's be honest, smartphones were the motherload. A TV, phone, and computer you always keep on you? They were salivating that day.

[–] SendMePhotos@lemmy.world 18 points 1 day ago (2 children)

Obscuration is what you're thinking and it works with things like adnauseun (firefox add on that will click all ads in the background to obscure preference data). It's a nice way to smear the data and probably better to do sooner (while the data collection is in infancy) rather than later (where the companies may be able to filter obscuration attempts).

I like it. I am really not a fan of being profiled, collected, and categorized. I agree with others, I hate this time line. It's so uncanny.

[–] HelloRoot@lemy.lol 3 points 1 day ago (1 children)

I still don't really understand adnauseum. What is the difference in privacy compared to clicking on none of the ads?

[–] SendMePhotos@lemmy.world 2 points 1 day ago (1 children)

Whatever data profile they already have on your can be obscured to make it useless vs them probably trickling in data.

Think of it like um...

Having a picture of you with a moderate amount of notes that are accurate, vs having a picture of you with so much irrelevant/inaccurate data that you can't be certain of anything.

[–] HelloRoot@lemy.lol 5 points 1 day ago* (last edited 1 day ago) (1 children)

But the picture of me they have is: doesn't click ads like all the other adblocker people (which is accurate)

Why would I want to change it to: clicks ALL the ads like all the other adnauseum people (which is also accurate)

[–] JustinTheGM@ttrpg.network 2 points 1 day ago (1 children)

They build this picture from many other sources besides ad clicks, so the point is to obscure that. Problem is, if you're only obscuring your ad click behavior, it should be relatively easy to filter out of the model.

[–] HelloRoot@lemy.lol 1 points 1 day ago* (last edited 1 day ago) (1 children)

You are just moving the problem one step further, but that doesn't change anything (if I am wrong please correct me).

You say it is ad behaviour + other data points.

So the picture of me they have is: [other data] + doesn’t click ads like all the other adblocker people (which is accurate)

Why would I want to change it to: [other data] + clicks ALL the ads like all the other adnauseum people (which is also accurate)

How does adnauseum or not matter? I genuinely don't get it. It's the same [other data] in both cases. Whether you click on none of the ads or all of the ads can be detected.

As a bonus, if adnauseum would click just a couple random ads, they would have a wrong assumption of my ad clicking behaviour.

But if I click none of the ads they have no accurate assumption of my ad clicking behaviour either.

Judging by incidents like the cambridge analytica scandal, the algorithms that analyze the data are sophisticated enough to differentiate your true interests, which are collected via other browsing behavious from your ad clicking behaviour if they contradict each other or when one of the two seems random.

[–] Ulrich@feddit.org 3 points 1 day ago (1 children)

[other data] + clicks ALL the ads like all the other adnauseum people

adnauseum does not click "all the other ads", it just clicks some of them. Like normal people do. Only those ads are not relevant to your interests, they're just random, so it obscures your online profile by filling it with a bunch of random information.

Judging by incidents like the cambridge analytica scandal, the algorithms that analyze the data are sophisticated enough to differentiate your true interests

Huh? No one in the Cambridge Analytica scandal was poisoning their data with irrelevant information.

[–] HelloRoot@lemy.lol 1 points 1 day ago* (last edited 1 day ago) (1 children)

adnauseun (firefox add on that will click all ads in the background to obscure preference data)

is what the top level comment said, so I went off this info. Thanks for explaining.

Huh? No one in the Cambridge Analytica scandal was poisoning their data with irrelevant information.

I didn't mean it like that.

I meant it in an illustrative manner - the results of their mass tracking and psychological profiling analysis was so dystopian, that filtering out random false data seems trivial in comparison. I feel like a bachelor or master thesis would be enough to come up with a sufficiently precise method.

In comparison to that it seems extremely complicated to algorithmically figure out what exact customized lie you have to tell to every single inidividual to manipulate them into behaving a certain way. That probably needed a larger team of smart people working together for many years.

But ofc I may be wrong. Cheers

[–] Ulrich@feddit.org 2 points 1 day ago (1 children)

filtering out random false data seems trivial

As far as I know, none of them had random false data so I'm not sure why you would think that?

In comparison to that it seems extremely complicated to algorithmically figure out what exact customized lie you have to tell to every single inidividual to manipulate them into behaving a certain way. That probably needed a larger team of smart people working together for many years.

I feel like you're greatly exaggerating the level of intelligence at work here. It's not hard to figure out people's political affiliations with something as simple as their browsing history, and it's not hard to manipulate them with propaganda accordingly. They did not have an "exact customized lie" for every individual, they just grouped individuals into categories (AKA profiling) and showed them a select few forms of disinformation accordingly.

[–] HelloRoot@lemy.lol 1 points 1 day ago* (last edited 1 day ago) (1 children)

Good input, thank you.

As far as I know, none of them had random false data so I’m not sure why you would think that?

You can use topic B as an illustration for topic A, even if topic B does not directly contain topic A. For example: (during a chess game analysis) "Moving the knight in front of the bishop is like a punch in the face from mike tyson."

There are probably better examples of more complex algorithms that work on data collected online for various goals. When developing those, a problem that naturaly comes up would be filtering out garbage. Do you think it is absolutely infeasable to implement one that would detect adnauseum specifically?

[–] Ulrich@feddit.org 0 points 23 hours ago

You can use topic B as an illustration for topic A

Sometimes yes. In this case, no.

Do you think it is absolutely infeasable to implement one that would detect adnauseum specifically?

I think the users of such products are extremely low (especially since they've been kicked from Google store) that it wouldn't be worth their time.

But no, I don't think they could either. It's just an automation script that runs actions the same way you would.

[–] dodgeflailimpose@lemmy.zip 1 points 1 day ago

Did not know this existed. I like the concept

[–] fubbernuckin@lemmy.dbzer0.com 8 points 1 day ago* (last edited 1 day ago)

I don't know if there's a clean way to do this right now, but I'd love to see a software project dedicated to doing this. Once a data set is poisoned it becomes very difficult to un-poison. The companies would probably implement some semi-effective but heavy-handed means of defending against it if it actually affected them, but I'm all for making them pay for that arms race.

[–] a14o@feddit.org 12 points 1 day ago (1 children)

It's a good idea in theory, but it's a challenging concept to have to explain to immigration officials at the airport.

[–] WalnutLum@lemmy.ml 2 points 22 hours ago* (last edited 22 hours ago)

"it says here you clicked 'sign me up for ISIS' 10000 times?"

"Haha no officer, you see it was my social chaff AI that clicked it"

[–] Ulrich@feddit.org 7 points 1 day ago

I have been a longtime advocate of data poisoning. Especially in the case of surveillance pricing. Unfortunately there doesn't seem to be many tools for this outside of AdNauseum.

[–] chonkyninja@lemmy.world 2 points 21 hours ago

There’s plenty of tools already, that can create many profiles of you, each with complete different personalities and posts.

[–] Core_of_Arden@lemmy.ml 5 points 1 day ago

So, she is talking about an AI-war? Where those who don't want us to be private, controls the weapons? Anyone else see a problem with that logic?

Thousands of "you" browsing different sites, will use an obscene amount of power and bandwidth. Imagine a million people doing that, not a billion... That's just stupid in all kinds of ways.

[–] wise_pancake@lemmy.ca 8 points 1 day ago (1 children)

In a different direction now is a good time to start looking at how local AI can liberate us from big tech.

[–] dodgeflailimpose@lemmy.zip 2 points 1 day ago

Local AI requires Investments in local compute power which sadly is not affordable for private users. We would need some entity that we can trust to host. I am happy to pay for that

[–] relic4322@lemmy.ml 2 points 23 hours ago

Ok, got another one for ya based on some comments below. You have all the usual addons to block ads and such, but you create a sock-puppet identify, and use AI to "click" ads in the background (stolen from a comment) that align with that identity. You dont see the ads, but the traffic pattern supports the identity you are wearing.

So rather than random, its aligned with a fake identity.

[–] regedit@feddit.online 5 points 1 day ago (2 children)

I did this with period trackers. I'm male and my wife and I would always chuckle when my period was about to start.

[–] Thorned_Rose@sh.itjust.works 2 points 22 hours ago

Drip. Don't give companies menstrual data at all ☺️

[–] dodgeflailimpose@lemmy.zip 3 points 1 day ago (1 children)

Like pretending to be menstruating? Or as a joke?

[–] regedit@feddit.online 1 points 21 hours ago

To fuck with their metrics and tracking attempts for legit female menstrual cycles.

[–] relic4322@lemmy.ml 6 points 1 day ago (1 children)

This is like chaff, and I think it would work. But you would have to deal with the fact that whatever patterns it was showing you were doing "you would be doing".

I think there are other ways that AI can be used for privacy.

For example, did you know that you can be identified by how you type/speak online? what if you filtered everything you said through an LLM first, normalizing it. Takes away a fingerprinting option. Could use a pretty small local LLM model that could run on a modest local desktop...

[–] dodgeflailimpose@lemmy.zip 3 points 1 day ago

I really like this idea

[–] edel@lemmy.ml 4 points 1 day ago

First, Naomi and her team are doing a fantastic work in security for masses, easily top 5 worldwide!

AI is capable but we are still failing at program it properly, gosh, even well funded companies are still doing a poor job at it... (just look at the misplaced ads and ineffective we still get.)

What I want, and it is easy to do TODAY, is AI checking our FOSS... so many we use and just a tiny, tiny minority of it goes with some scrutiny. We need AI to go through the FOSS code looking for maliciousness now.

[–] Reverendender@sh.itjust.works 4 points 1 day ago (2 children)

It’s an interesting concept, but I’m not sure the payoff justifies the effort.

Even with AI-generated noise, you’re still being tracked through logins, device fingerprints, and other signals. And in the process, you would probably end up degrading your own experience; getting irrelevant ads, broken recommendations, or tripping security systems.

There’s also the environmental cost to consider. If enough people ran decoy traffic 24/7, the energy use could become significant. All for a strategy that platforms would likely adapt to pretty quickly.

I get the appeal, but I wonder if the practical downsides outweigh the potential privacy gains.

[–] HelloRoot@lemy.lol 6 points 1 day ago* (last edited 1 day ago) (2 children)

getting irrelevant ads

you guys are getting ads?

[–] edel@lemmy.ml 2 points 1 day ago

My entire family is ad free for years... with the exception in podcasts. I am tempted to block them too (is there a way now?) but still not too intrusive... it is a way for me to keep connected to the ad world anyways. Now, the moment they abuse them here tii... I'll find a way to block these.

[–] Reverendender@sh.itjust.works 2 points 1 day ago (2 children)

I’m not, but OP would if they started opening up their IP and fingerprints to anyone who wants them, in order to inundate those parties with garbage data. Admittedly, I might be missing some clever part of their plan.

[–] blackbrook@mander.xyz 4 points 1 day ago (1 children)

Getting more targeted ads is not really in your interest. That is an idea promoted by the ad people.

[–] Reverendender@sh.itjust.works 0 points 1 day ago (1 children)

I’m not seeing the relevance of your comment

[–] blackbrook@mander.xyz 1 points 18 hours ago* (last edited 18 hours ago) (1 children)

You said

you would probably end up degrading your own experience; getting irrelevant ads

Irrevant ads = less targeted ads. You seem to think this is a negative. I'm saying it is actually a positive.

[–] Reverendender@sh.itjust.works 1 points 18 hours ago

Taken by themselves, yes you are correct. Less targeted is a positive. I would rather just not have the ads, which is where I am at now. So I don’t understand why I would give that up, which is what the OP’s strategy would seem to require.

[–] dodgeflailimpose@lemmy.zip 2 points 1 day ago

No clever plan. Just picked up this idea and like to see different opinions from people maybe far more advanced in that field

[–] fubbernuckin@lemmy.dbzer0.com 2 points 1 day ago (1 children)

Okay but irrelevant ads is the dream. I'd prefer not to get recommendations at all either. I'll hear from word of mouth what's worthwhile to watch, or I'll look for it myself. Recommendations consistently muddy things up, it makes all modern social media useless, I have no idea how people can put up with it.

[–] Reverendender@sh.itjust.works 1 points 1 day ago

I agree, which is why this approach to me seems ultimately counterproductive on an individual level.

[–] slackness@lemmy.ml 2 points 1 day ago

You would be able to do this for a short while but unless you can make an agent that's indistinguishable from you or you already have very bot-like traffic, they'd catch up pretty quickly. They aren't going to just let a trillion dollar industry die out because some bots are generating traffic.