The danger of governments using AI to target proxy

[torikki-tou]

A few weeks ago an internal document of Roscomnadzor (Russian internet-oriented law enforcement agency) was leaked. I found a screenshot of that document that caught my attention.

Research and development work was organized on:

• promising encryption technologies and protocols: Trojan, Vmess, VLESS, V2Ray; prepared proposals to counteract their use
• the possibility of identifying long-term information influence operations
• the possibility of transcribing musical works to identify signs of violation of the legislation of the Russian Federation

Research was conducted:

• to study threats of software for cyber espionage (Pegasus)
• on the use of artificial intelligence technologies to identify traffic of VPN and proxy-based solutions

We are interested in the first and last points.

The first point is not something new, of course they have been investigating possible vulnerabilities of Xray and its fellow protocols. But I wanted to point out and hopefully to start a discussion on the last line, where they state that they will be using AI to target proxies and VPNs. I see two possible attack vectors:

1. They can try to analyze traffic fingerprints with AI using DPI or something similar. XTLS Reality claims to be undetectable but have somebody tried to look it up with AI tools? Is there any understanding whether it can be detected that way and if it even makes sense? I would love to hear from Xray-core devs on this.
2. They can try to analyze traffic distribution between servers and clients to detect proxies and their clients. It seems more dangerous. I heard in Iran its a usial thing for VPS to get banned after X Gb of traffic. Is it only CDNs who can save us in this case?

[RPRX]

AI 可以检测出所有的代理协议，宏观特征太多了

[torikki-tou]

AI can detect all proxy protocols, there are too many macro features

So what do we do?

[zertmark]

So what do we do?

I guess we need to use AI ourselfves to mask (Jesus, save us all)