Cider9986

You should threat model, but if you want to maximally protect yourself, use Vanadium or Brave Browser and GrapheneOS and no privacy invasive apps.

Going by this definition [1]:

>Privacy is the assurance that your data is only seen by the parties you intend to view it.

I’ve been assuming we’re using open source software. When developing open source, you’re putting your code out publicly. Your code is not private data.

As far as the agent acting on your computer, you can easily use Claude Code on the web on Anthropic’s VMs, or just run it in something like UTM or another user on your laptop.

So I don’t see how this is a big loss of privacy. By the definition, you’re intending Anthropic to see your code just as anyone else can.

Ways you could use AI in a non-privacy respecting way would be like letting it read all your personal notes while assuming the AI company will not share them or look at them. I don’t see code as sensitive like this.

Your prompts, pattern of speech, usage patterns, when you work, stuff you say to the LLM besides coding related. Those would be a fair argument for compromising the privacy of the developer. But as long as you consider what you say to be public, then you're intending that data to be out their, so your privacy isn't being compromised. You could make the same argument about developing in the open on GitHub. Anyone can learn about your usage patterns, time zone, etc. It is also fair to say you're giving that AI company more power over you by giving them exclusively that information, while on GitHub it would be information given to everyone. .