I’m not sure if it is entirely accurate to compare them in this way, as “Matrix” refers to simply the protocol, whereas “Signal” could refer to the applications, server, and protocol. That being said, is there any fundamental difference in how the Matrix ecosystem of federated servers, and independently developed applications compares to that of Signal that would make it less secure, overall, to use?
The most obvious security vulnerability that I can think of is that the person you are communicating with (or, conceivably, oneself, as well) is using an insecure/compromised application that may be leaking information. I would assume that the underlying encryption of the data is rather trustworthy, and the added censorship resistance of federating the servers is a big plus. However, I do wonder if there are any issues with extra metadata generation, or usage tracking that could be seen as an opsec vulnerability for an individual. Signal, somewhat famously, when subpoenaed to hand over data, can only hand over the date that the account was created, and the last time it was used. What would happen if the authorities go after a Matrix user? What information about that user would they be able to gather?
In what way is it watered down and what are the benefits?
Two ways mainly: the private key is no longer device specific but rather account specific and the ratchet required for perfect forward secrecy is not forwarded on every message AFAIK.
The benefits are better scalability in larger group chats and generally less key exchange hassles. But this comes at a significant downgrade in security.
Basically MegOLM is only slightly better than OpenPGP / OTR or similar older generation e2ee systems.
Not to downplay the security of OLM, but if MegOLM is as cryptographically sound as OpenPGP, then that’s already very secure.
As long as my private key is safe, it would be computationally impossible for anyone to decrypt messages intended for me.
Yes, the problem with MegOLM however is that the private key is automatically exchanged between devices using the same account. I think 2-3 years ago there was an exploit allowing malicious home-server admins to add additional devices under their control and thus listen in on what appeared to be e2ee conversations to the user. This specific exploit was fixed, but the principal problems of potentially insecure exchange of private keys remains in MegOLM to this day.
OpenPGP is not very secure, by modern standards.