I once did this very thing, with the added feature of being able to send messages to either user (masquerading as the other). It was fun as a proof of concept and as a way for me to learn Python Twisted, but using this technique for an extended length of time is a bad idea for various reasons.

Edit: one example of things going bad for you:

Stranger 1: 13/f/fl, u?

Stranger 2: 18/m/NY. Can I send u pics?

Congrats, your IP is a couple of bad decisions away from committing a felony.

This is my first Coffeescript project. This has already been done before, but I didn't bother to Google before inventing it, so here it is. You can specify interests that strangers will talk about.

I've done this before with perl and POE::Component::Omegle. IIRC a large majority of the conversations were insta-disconnects, and those that lasted longer than that were very, very boring.

I guess actually using this program would be blatantly illegal. But it's an interesting problem--how to defend against this? One way is to compute a fingerprint of the Diffie-Hellman shared secret, and then have the chat participants vocally compare it (I think one of the "secure chat" programs do this?). That should be secure as long as the attacker can not imitate voices, but it's rather heavy-handed.

What you describe is standard ZRTP. You don't really have to imitate voices if it's someone the participants have never ever seen before, but you do have to get the lipsync right. Does Omegle have sound?

>That should be secure as long as the attacker can not imitate voices,

EDIT: Messed up reading. I think the major problem with this idea is that it makes the barrier to entry for using omegle much higher, suddenly you need a microphone.

Original post below the cut:

---

They can.

https://news.ycombinator.com/item?id=10020701

https://news.ycombinator.com/item?id=10020710

Hm, did you paste the right URLs? I don't see how they are related to this thread.

They seem to point to the right place on my computer. The first should be a post I made about how this works, in the second somebody talks about doing it and how he could imitate peoples voices.

EDIT: OHHH, vocally. I did a misread ignore me.

I made of these back in the first month that Omegle was launched. I think they started automatically detecting it and quarantining these types of bots pretty quickly, though.

One solution I can see now, is to use something like word2vec, or better yet, skip-thought vectors, to replace each sentence with synonymous sentences and avoid tripping their detector.

Question "from the other room": how does this work? Does it come from exploiting some bug in Omegle code?

There was a program that could do this for years called omeglespy[0]. How it works is:

You make two connections to omegle, and are connected to two different strangers:

You take Stranger A's words and pass them off as your own to stranger B, and vice versa. From your perspective you're watching the entire conversation, from their perspective they're talking to each other without interference.

[0]: https://code.google.com/p/omegle-spy/

That graph is very misleading. It's not a real MitM attack, is it?

It is a MITM. Unknown key share would be the better definition I believe. Alice connects to Eve, Bob connects to Eve. Eve passes what Alice is telling Eve to Bob, and Bob to Eve. Alice thinks she's talking to Bob.

Alice doesn't know who Bob is, though.

Well she's confused. She thinks what Eve is showing Alice is Eve, but it's actually a copy of Bob.

How is this even remotely ethical?

When chatting on Omegle, you're paired anonymously with an anonymous stranger, with no guarantee of whether that stranger will record anything or pretend to be someone else. I think the front page http://www.omegle.com/ makes that abundantly clear.

Same level of ethicality as opening two Omegle windows and forwarding messages between players.

Who claimed it was ethical?