Wanted to link @pulls comment from another thread discussing DAITA v2
How common are AI traffic analysis tools nowadays?
I’m not educated on this topic, but last time I read up on this, they were mostly proof-of-concept research papers written by computer scientists in Chinese universities (for VPN and Tor connections.
Then again, perhaps things have indeed changed for the worst 
There have been a lot of cases where Tor users have been deanonymized by LE, and the method mentioned is usually a vague term like “traffic analysis”. So I would say very common.
Example: Law enforcement agencies undermine Tor anonymisation
ML traffic analysis is presumably what would be used to exploit netflow (or similar) data for deanonymizing VPN traffic. The fact that the US is spending many resources for purchase of VPN traffic from ISPs, is a strong indicator that such ML traffic analysis tools are already used.
Surely, Western special services like FBI NSA would use these tools, and they could purchase bulk traffic from major ISPs that host the autonomous systems where the majority Tor nodes are on. So they wouldn’t even have to own the Tor nodes to get some success at de-anonymizing users. But I strongly doubt this would be “common” amongst less powerful governments, local police forces or companies.
ML has long been used for traffic analysis & correlation. The “AI” moniker seems like a marketing gimmick device, as I don’t think NLP models (like LLMs) that people have come to know as “AI” are in anyway related.
Could have just as well named it “protects from ML threats”, but that’s missing the hype train: For businesses that rely on marketing, fake/misleading/half-truth or not (like VPNs)… not doing so is a self-own.
DAITA does not appear to be automatically enabled. Mullvad VPN users can go to Settings> DAITA (on or off).
Took me a minute to find out if it was built into the release and automatically enabled just simply by using Mullvad VPN or a user selected option, so I thought I would share.