If you have interoperability, then you need cipher negotiation between clients with different capabilities (and they will always have different capabilities), and that's a huge, juicy attack surface. Multiple critical SSL/TLS CVEs-- including some we know for a fact the NSA relied on-- came from cipher negotiation.
> If you have interoperability, then you need cipher negotiation between clients with different capabilities (and they will always have different capabilities), and that's a huge, juicy attack surface.
Not really. The degree of malleability in cipher negotiation is widely considered to have been a Bad Move in SSL/TLS's early design, and modern (well-designed) cryptographic protocols don't enable the kinds of parametric malleability that made SSL/TLS so exploitable at the time.
Signal's protocol, for example, is perfectly interoperable; the lack of interoperability comes from a (not unreasonable) constraint at the application layer, not the protocol itself. Another example would be MLS[1], which supports fixed suites rather than parametric malleability and uses the technique from RFC 8701[2] to prevent clients from getting clever and trying to add their own extensions that undermine the fixed suites.
