As far as I know, even when you use hidden services it is enough for the 1st and the 3rd to be malicious in order to de-anonymise a user. The tor security is barely enough (if it is enough at all) - there is no reason to believe that the FBI/FSB/etc don't have enough relays up in order to de-anonymise most users. I2P is much better in that regard.

Heck, until recently the tor team used 80-bit truncated sha-1, 1024-bit rsa, and 128-bit AES for their traffic, not to mention that the tor browser ships with javascript enabled by default.

In theory, I don't see how entry and exit are enough to deanonymize. They don't even know that they're in the same circuit unless they have a covert channel (like relay early) or manage traffic correlation during the ~10 minute circuit lifetime.

And then, using onion services, there are two three-relay circuits that meet at a rendezvous point. One picked by the onion, and the other by the user. So even deanonymizing one of those circuits would be insufficient.

But that's all theoretical. In practice, there are likely undisclosed vulnerabilities. Perhaps lots of them.

I do agree that the Tor browser standalone is rather a joke. Especially if it's in Windows. You at least want to be using Whonix. And if you really care, Whonix in Qubes.

> They don't even know that they're in the same circuit unless they have a covert channel (like relay early) or manage traffic correlation during the ~10 minute circuit lifetime.

No reason to think that they would not do that.

> And then, using onion services, there are two three-relay circuits that meet at a rendezvous point. One picked by the onion, and the other by the user. So even deanonymizing one of those circuits would be insufficient.

It would be sufficient to de-anonymise one of the parties.

> In practice, there are likely undisclosed vulnerabilities. Perhaps lots of them.

My point is that you do not need an undisclosed vulnerability to break tor if you have enough resources.

One major threat factor that Tor doesn't have a bulletproof solution to, and likely never will, is correlation attacks. It's been shown to be plausible that by observing the timing and size of packets, even without knowing the contents, is enough to determine that two relays are part of the same circuit.

Well, any system that relies on tunneling is vulnerable to correlation attacks. And drilling down by looking at traffic between autonomous systems. Unless it uses chaff to maintain constant throughput.

But will doing that be worth it to find someone like me? I doubt it. I'm just a hobbyist and writer.

As far as I know garlic tunnelling that i2p uses is not as vulnerable.