The Privacy Dilemma.

Everyone wants your data.

OpenAI wants it. Google wants it. Anthropic wants it. Meta wants it. Your government wants it.

Especially your Agent Traces -- the most valuable kind of data out there.

It has never been harder to maintain a level of Privacy in the age of AI, and there's a simple reason why.

Perverse incentives.

Think about it -- a company that sells ads (Google, Facebook) has a very strong incentive to gather as much data about you as possible, in order to make more money with their ad selling business.

They are literally incentivized to invade your privacy and gather as much info about you as possible. They could not care less about your privacy.

This has been common knowledge for a while, however what has changed is that there are now two new major players in this equation.

OpenAI and Anthropic.

And that our agent traces contain more sensitive & private data about us than ever before.

These two AI labs do not make most of their revenue via Ads (although OpenAI is flirting with that idea for free ChatGPT users... which means you have to pay to get any level of privacy).

The primary way these two AI companies make money is inference. They sell tokens.

Both Anthropic and OpenAI are racing to make the best AI model possible, because whoever has the best model sells the most tokens --> makes the most revenue.

But ask yourself: How does an AI lab make the best model? Sure, you need compute. Sure, you need some talent -- but the main thing you need is Data.

Without loads of high-quality data, you cannot train a frontier LLM.

By using pure logic, we can quickly realize that the incentives for OpenAI and Anthropic aren't much different from Google and Meta. They want your data. All of it.

Especially your agent traces.

It's no secret these companies are training on your data. Recently, there's been a story of OpenAI using similar techniques to achieve some of its Math breakthroughs, as some professional mathematicians using ChatGPT.

Did they train on their chats? Did they specifically target those scientists by giving them free ChatGPT Pro $200 plans? Did they "borrow" the same line of reasoning to achieve these breakthroughs?

We don't know for certain. But one thing we do know is that OpenAI and Anthropic want more and more of our data -- the most sensitive and private information we have, fed into Claude and ChatGPT.

So what is the solution then?

Decentralize. You must find software & companies that have incentives aligned with you.

Instead of using Google Docs, use Obsidian. Instead of using WhatsApp, use Signal. Instead of using Cursor, use Cloudroom.

Ask yourself: How does this company make money?

Google and Meta make money by selling your data to advertisers. Anthropic and OpenAI make money by training new models on your data.

You must diversify your software stack, so that for each purpose you are using the best aligned company with your goals.

Better to have all your data spread out across 8 different companies, than just two big-tech giants.

A company that cares about Privacy:

1. does not sell your data to third-parties
2. does not train AI models on your data
3. is open-source

This is why a lot of people are switching to Cloudroom. Power users care about privacy, and Cloudroom does not train any models, it does not run ads, it does not sell data to third parties, and is fully open-source.

And since your agent traces are some of the most important data you will have, you should choose the platform you're running your agents on carefully.

Choose software that is aligned with you.