citation needed

It doesn’t have the keys, but it can inject any javascript and do whatever the user can do.

To me it seems they've taken all precautions they can reasonably take -- "what if the user installs a keylogger" isn't fixable by anyone.

"Installing a keylogger" is a vast oversimplification, even if it is outside of their threat model.

Installing almost anything in your browser is usually a matter of a couple of clicks.

Like intercepting your oauth token next time you login into SSO and then use that to access your tailnet.

This was true even before this new feature.

The new threat model is entirely psychological.

Sounds like about "as good as this gets" if you happen to want to do this in browsers. Good job.

And then the addon intercepts the loading of the Wasm code, injects it’s own payload into it and has access to the keys.

What keys? I think the implementation does not use regular SSH keys for SSH authentication, but rather something custom (I believe traffic to port 22 on each SSH enabled client is intercepted and the daemon handles authentication itself).

For the downvotes: Am I misunderstanding Tailscale's implementation?

I'm not commenting on whether it's a good or bad idea, but we should at least be talking about the same thing.

The wasm code runs in the browser, hence the keys live in browser memory, so the browser has access to the keys.