AI’s Nature of Surveillance

With the emergence of consumer friendly AI agents over the past few years, there is a great deal of excitement around using these tools that will “replace workers, make your tasks easier, help code the technology of the future, destroy humanity” etc. With all of this excitement, it’s important to understand the privacy issues when using AI. Artificial Intelligence and their modals are a huge surveillance apparatus. AI models are different programs that can detect patterns using a specific dataset. There are competing organizations that release their own models that use different algorithms and datasets.

How AI Works?

AI models are trained on large amounts of data. The hardware best optimized for training these models are Graphical Processing Units (GPUs). Much like Bitcoin mining operations, there’s been a growth of data centers focused around GPU hardware. GPUs are better suited for processing data than CPUs. Over a certain period of time, the AI is optimized to better fit the parameters of its training data. Then the model is deployed for people to use. The AI models learn further from user input.

AI gets better and more refined by ingesting more and more data

Enter Privacy Issues

As mentioned above, AI models are trained with large data sets, and with more data, the AI can be better refined. Data that is used for training by the AI companies includes anything available at their fingertips. After the AI model is made available to the public, the user input data is also used to train the AI. Your online data and your inputs are used to train these models. The more information that can be fed into the AI, the better equipped the AI becomes. Big data is what AI and the companies around AI are looking for.

How do you think Meta trained their model? They used their social media platforms and EVERYTHING posted on their since at least 2007.
Another example is Grok (The AI of x.com) is being trained on users’ tweets. Consider anything being posted to the open internet is up for grabs. Much like how you may be concerned about regular digital privacy concerns when posting online, AI looks to collect everything it can access and build better models using this same data.

https://videopress.com/v/4qZHr9iK?resizeToParent=true&cover=true&preloadContent=metadata&useAverageColor=true

Peter Thiel describes the AI hunger for more and more data, and how it trends towards centralization. “AI is about big data and big governments controlling all the data”

AI to your advantage

AI is a powerful tool and you can still wield it to help make your life a bit easier. There are a few things you should take into account when using an AI model:

Do you need an account to use the service?
Can you use it over Tor?
Can you pay in Monero or some other crypto?
Is the data stored?
Is data encrypted client-side?
Can it be run locally?
Can you self-host locally?
Are there any censorship/safety features in place?

There are different platforms that offer access to AI models with a variety of different obfuscation methods. Some include more than others. If you understand how AI models work, you are better prepared in creating prompts and using AI without giving it a treasure trove of data about the user (YOU).

Private AI Models:

Here are a few options including the benefits of each.

Duck.ai (https://duckduckgogg42xjoc72x3sjasowoarfbgcmvfimaftt6twagswzczad.onion/?q=DuckDuckGo+AI+Chat&ia=chat&duckai=1)

This onion address can be used over Tor
Duck.ai acts as a proxy between you and the AI
This includes GPT4o-mini, Llama 3.3 70B, Claude 3 Haiku, o3-mini, and Mistral Small 3
Duck.ai’s terms of service also states that the prompts cant be used to train the AI models and that the data will be deleted after 30 days
There are no accounts needed to use duck.ai

https://venice.ai

Can be accessed over Tor
Can be used without an account
Includes text, image, coding specific and uncensored models (these are the venice.ai guest models, there are more models available for a PRO account):
- Text: Llama 3.2 3B, Llama 3.3 70B
- Image: Fluently XL Final, FLUX Custom, FLUX Standard, Pony Realism
- Code: DeepSeek Coder V2 Lite, Mistral Codestral 22B, Qwen 2.5 coder 32B
Can upload PDFs and TXT documents
Data is stored in the browser and is encrypted when in transit
Doesn’t use data to train models
Pay for PRO with bitcoin

https://jan.ai

FOSS application
Can run offline
Data is stored locally on your device
No account
Access to MANY different models and can import other models

https://devinci.onicai.com/

No account needed
Entire session runs within your browser session
- AI models are downloaded into your browser cache
App can be installed and run offline
Access to many different models

https://nano-gpt.com

No account needed
Can be accessed with Tor
Data is stored on your device
pay per prompt with Monero or other crypto (discount for nano)
Can gift others access
Access to many different models including text, image and uncensored agents

https://ppq.ai/

No account needed
Can be accessed with Tor
Data is stored on your device
pay per prompt with Monero or other crypto (discount for lightning)
Can gift others access
Access to many different models including text, image and video models including text to video agents

https://cryptotalks.ai/

pay in Monero or bitcoin
Need to create an account
Can be accessed with Tor
Access to many different models including text, image, and uncensored models

https://gpt4all.io

FOSS application
Can run offline
Data is stored locally on your device
No account
Access to MANY different models and can import other models

You can find all of our AI resources located at https://sovereignstack.tools/privacy-resources/ under Private AI Models.

Final Considerations

When using normal AI agents on platforms that aren’t privacy preserving, there are a few things to take into consideration. The more information you give freely to the AI systems, the more detailed the picture they can build about you. Even if you use the AI sparingly, given enough pieces of information, the AI will be able to fingerprint you. These AI agents and the websites around them are surveillance machines, their goal is to collect as much info about you as possible. they’re hungry for information, they are grabbing your device fingerprint, timezone, how you move the mouse on the website. Just check out https://amiunique.org to see how just a few different attributes make you completely unique, then when you add details about yourself, you really help build a surveillance profile about yourself. Avoid them at all costs.

Use a VPN or Tor, don’t give identifying information if you can help it, don’t upload sensitive documents to these models, and use the private solutions above. There will be more privacy developments, we must embrace and support these tools that stand against the surveillance apparatus.