arXiv:2610.16100v2 Announce Type: replace Abstract: A deniable authenticated key exchange (DAKE) establishes an authenticated shared key without giving third parties proof of participation. Post-quantum DAKEs use ring signatures, message authentication codes, or split key-encapsulation mechanisms, which differ in who can authenticate and how deniability is achieved. We introduce \emph{authentication allocation} to classify which parties hold secrets that produce accepted authentication messages. For protocols using a fixed two-member ring, we prove that an accuser can simulate a session by either sharing the accused party's authentication key or signing with its own ring key. Thus, deniability relies on key sharing or the ring signature's ability to hide its signer. We propose \HDKEM, combining ML-KEM-768, X25519, and an optional pre-shared or quantum key distribution (QKD)-delivered key $k_3$ through a split-key pseudorandom function. The session key remains secret if at least one input remains secure. Since provisioning services and QKD key managers may know $k_3$, the client always authenticates with its own ring key; the mode determines whether other holders of $k_3$ can impersonate the server. Four modes are supported: \emph{source-backed} authenticates the server through key confirmation under $k_3$, allowing its holders to impersonate the server; \emph{hide} also encrypts the client's signature; \emph{mutual} adds a server ring signature to prevent such impersonation; and \emph{source-less} authenticates both endpoints with ring signatures without $k_3$. We prove extended Canetti--Krawczyk (eCK) session-key secrecy in the quantum random-oracle model, explicit authentication, forward secrecy, and session-partner deniability, with a concrete deniability bound in the classical random-oracle model. With FalconRS, source-backed mode sends 3,813~bytes in about 26~ms, versus 5,101~bytes and 47~ms in mutual mode.