Summary
A flaw in TLS stateful session resumption allows cached sessions to be reused under a different SNI/ALPN than originally negotiated, bypassing intended binding checks.
In environments where client-authentication policies vary by virtual host or protocol, this can let an authenticated state leak into a context for which it was never established, weakening access controls.
Affected context
This vulnerability arises in TLS deployments that:
- Use stateful (session-ID) resumption alongside multiple SNI-based virtual hosts or ALPN-negotiated protocols
- Enforce different client-authentication or authorization policies per virtual host or protocol
- Rely on the assumption that a resumed session is tightly bound to the original SNI/ALPN context
Under these conditions, a cached session may be resumed in a different logical context than the one where its authentication state was originally negotiated.
Exploit preconditions
- Attack surface: TLS endpoints offering stateful session-ID resumption with multiple SNI/ALPN contexts.
- Required access: Ability to act as a legitimate TLS client and initiate or resume sessions against the server.
- Trigger point: Initiating a session resumption using a cached session while presenting a different SNI or ALPN than was used in the original handshake.
- Why exploitation is possible: The resumption logic did not verify that the cached authentication state was bound to the same SNI/ALPN, allowing peer-authentication to be reused in an unintended context.
Why this happens
The underlying issue is an authentication logic flaw where the server's stateful TLS resumption mechanism fails to bind session state to the specific endpoint identity and protocol (SNI/ALPN) for which it was created.
This aligns with CWE-287: Improper Authentication, because the server treats a resumed session as valid authentication without confirming that the resumed context matches the original security policy.
Abstractly, any mechanism that caches authentication state but does not enforce strict context binding (hostname, protocol, or policy realm) risks leaking trust across boundaries and enabling privilege use outside the intended scope.
Potential impact
- Authenticated client state may be reused on a virtual host with different client-authentication requirements.
- A client could gain access to resources on a host or protocol where it should not be considered authenticated.
- Policy separation between SNI-based virtual hosts or ALPN-negotiated services can be weakened or bypassed.
- Regulatory or compliance controls relying on strict host-level client-authentication may be undermined.
General mitigation guidance
Effective mitigation focuses on enforcing strict binding between authentication state and its intended context.
Common patterns include:
- Ensuring all session resumption paths (stateful and ticket-based) validate SNI and ALPN consistency before reusing authentication state.
- Designing authentication caches so that stored credentials or session IDs are keyed by explicit context identifiers (e.g., hostname, protocol, policy realm).
- Applying least-privilege principles to per-virtual-host policies, avoiding broad, shared authentication state across hosts or services.
- Treating resumption failures or mismatches as signals to perform a full handshake and re-evaluate authentication under the correct policy.
How teams detect issues like this
Detection typically involves reviewing how TLS resumption and authentication policies interact:
- Static or configuration analysis to confirm that session caches are keyed and validated against SNI/ALPN or equivalent identity markers.
- Code review focusing on resumption paths to ensure policy checks are identical across ticket-based and stateful resumption mechanisms.
- Policy enforcement that requires explicit context binding for any reused authentication state, flagged during security architecture reviews.
- Use of automated security testing or protocol fuzzing tools to attempt cross-host or cross-protocol session resumption and observe unexpected authenticated access.

