They talk about decrypts of SSH tunnelling as a "potential", if they can later steal the keys. SSH, I note, does have an RSA-based key exchange as well as its usual Diffie-Hellman: if their targets have been using plain RSA, that would make an attacker's life easier for historical decrypts! Based on their typical methodology, I think that is
probably what they are talking about, because they mention stealing IPsec pre-shared keys from router configurations in a very similar context.
Of course, all we have is a probably - this selection of documents is not anywhere near as comprehensive as we'd perhaps like here. We're having to fill in the blanks - and there's too many blanks to fill in clearly. There seems to be relatively little in this leak from NSA's PICARESQUE/PIEDMONT, or GCHQ's STRAP3 (which covers specific operational details: purely by way of hypothetical example :-), where individual full-take feed taps actually are in Telehouse North, or specific details about SIGINT enabling via the Cavium Nitrox chips), sadly. Alternative ideas (or leaks!) are welcomed.
Recent versions of OpenSSH have using some non-NIST primitives from djb, including Curve25519-SHA256 key exchange, Ed25519 keys and ChaCha20-Poly1305 transports. I am quite confident neither NSA nor GCHQ have any good cryptanalytic attacks against those primitives.
There is mention of some exploitation against finite-field Diffie-Hellman in TLS (PHOENIX). That lacks context, however, and we can only guess about what's missing. One possibility is it's an active attack which tricks the peers into agreeing keys over an unsafe field (TLS 1.2 has no way to date for peers to suggest or agree on lists of named fields; however, the recent ffdhe draft does provide one) - however active attacks don't really fit the context under discussion in the slides. The TLS Working Group at IETF is currently discussing this, and a suggestion's been made to remove the old finite-field DHE transports due to their poor performance and apparent vulnerability, replacing them with ECDHE over secp256r1 (a NIST curve), and quite possibly in the near future X25519 over Curve25519 (a non-NIST curve). I don't know how that's going to resolve yet.