If you're a software architect or a software developer designing the architecture of future cloud systems or future server systems, I would seriously consider consistently using a no trust model where the server doesn't have access to the data it processes.
AI will eat anything weak for breakfast, and hack pretty much anything if someone is willing to foot the bill for the tokens.
With this new threat level, we should assume that everything in the cloud will eventually be hacked and/or exploited. The best way to protect data from a hacked system (whether you know that the server is hacked or not) is to make sure the system has access to as little useful information as possible. This is the opposite of today's common corporate model where companies want to slurp up as much data as possible to monetize it however possible. That has implications for how the companies we make the software for plan to make money in the future. Personally I like the idea of companies relating to the users of their software as customers - not as exploitable consumers.
A server and its software should only have the minimum information required to do whatever it's supposed to do. Any "nice to have" features that require more personal or secret data, like the decryption keys, personal information, contents of messages/emails, IP numbers, routing information, images etc., should simply be left out. Anything that can use end-to-end encryption should use it. For example, if you have a company with a thousand employees and the messaging server is hacked - the messages will still be secure if they are encrypted and the server doesn't have the keys. If a hundred of the employees' devices are hacked, the compromise can still be contained to the information available to those devices.
In other words, going forward we should be very mindful about what is stored in clear text or even encrypted - with easy access to the keys - on any device. If we assume that almost everything online eventually will be hacked - which I think is a reasonable assumption - we also need to be mindful about what we don't need to store online, and what we need to store off-line. Like backups and essential documents and documentation. And family photos and videos of your pets.
For example - today's popular cloud storage options in their normal mode; like Nextcloud, Dropbox, Google Drive, Microsoft's OneDrive can all potentially expose the contents of your files in a server-side compromise. Some of them, including Nextcloud, can use end-to-end encryption and prevent the storage servers from seeing the actual content of the files. In the future, full end-to-end encryption must be the default for any cloud storage - and it should be the only option for any new cloud storage systems we design.
Any politician who wants a "backdoor" into anything must be educated. The backdoors already used by the government, for example in telecom, can also be used by other unfriendly governments and criminals, like the Salt Typhoon incident demonstrates.
The same goes for messaging. Normally messaging servers don't need to see the content of messages. If we need features like search, we should consider caching all the relevant messages or the search index locally on the end-users' devices.
My own IM messenger DarkSpeak doesn't even use servers. That's the optimal solution for these kinds of threats - but it's only practical in very special use cases.
We can't prevent every device and every server from being hacked. What we can do is design systems where hacking one of them doesn't give the attacker everything.