Proton Meet Isn't What They Told You It Was

1. You’re right, i should have written: among the thousands of bugs or vulnerabilities that were reported found by Claude, none was an encryption crack. But it could have cracked one without it being reported. That said, to crack a cipher, you don’t do a “computer attack” like you would do a pentest. You do pure maths instead and Claude’s new model is not targetting particular maths skills.
2. I claim that the infrastructure is irrelevant for that specific fear because it is, and i know that it is because i have studied it and it’s my job. I cannot teach years of computer science within a post. But i thought the argument was clear enough: re-read the article in the OP and you will see that it all revolves on revealing IPs (repeated 100 times in different ways). IPs are not meant to be hidden from the infrastructure provider and IPs are easy to hide with a VPN (or a mere proxy, actually). So there is not even a real issue here. No other company will do more than what Proton does on this matter, and this for a simple reason: it would be very very expensive, and there is absolutely no point, no added value, no increase in privacy and no increase in anonymity if you use a VPN
3. You mention what you said in a group on Telegram. Using Telegram should worry you a lot more than hypothetical concealed AI breakthrough. AI is not magic. It’s not a secret big power in the hands of few illuminati, even if it’s true that it’s very powerful, therefore expensive, and therefore only billionaires (who seem to be incapable of being decent humans) can own it. It’s the result of thousands of mathematicians across the world publishing theoretical articles and reading their peers’ articles to build up upon them, and thousands of engineers trying to implement and test those theories. When Claude achieves spectacular results that surprise the public in media reports, tens of thousands of researchers and engineers were already knowing since several months what Claude would more or less do. And even if some might be bad persons, most of them are honest workers doing it for the art, the glory or science, not to implement dark schemes of world domination. This is how research and breakthroughs work. This was my job at some point. So if there was some breakthrough on breaking some major cipher, that would not stay a secret for long. It’s math disproven by many well-known researchers, not a zero-day vulnerability in a software. There is a long track record of ciphers being broken already. As far as I remember, every cipher was broken by researchers who published a paper about it.

Studying Proton technology doesn’t need to get their data through their infrastructure provider at all:

1. Proton products are open source, so you can just study the code on the repositories.
2. The data that is received/processed by the infrastructure and that is linked to a user is the (encrypted) data that leaves a user’s computer or arrives to a user’s computer. So just use a simple sniffer like wireshark or tcpdump while making a call with Proton Meet and you will have as much data as the infrastructure provider. Even if the goal is to identify patterns rather than people, an evil CEO of an evil AI company would litteraly just have to create many proton accounts and sniff the traffic on hosts during calls to feed its model with the data. No need to resort to an infrastructure provider.

Thank you for your advice on general logical thinking, I will try to apply it. But please allow me some advice too: I encourage you to study computer science or computer engineering during a few years in a university, to have some sense of what is possible, what is not, how pieces of tehnology work and assemble together, and what endeavour would make sense or be pointless for a given goal. That will refine your perceptions of patterns, trends etc. And if you do (“the choice is yours”), you will discover that, as for every topic scientifically analyzed, the “common sense” that had prevailed intuitively for years or even centuries was plainly wrong.