Explore our new Cybersecurity Terms Glossary, explained by our experts.

View the Glossary View the Glossary
PRIVIA

Cybersecurity Terms Glossary

What Is DPoP?

DPoP binds an OAuth access token to a key held by the client, so a stolen token is worthless without the private key that proves possession.

Illustration for the cybersecurity term DPoP
Home / Blog / Cybersecurity Terms / What Is DPoP?
17 November 2026

What Is DPoP?

What Is DPoP?

An ordinary OAuth access token is a bearer credential, and the word bearer is doing real work: whoever holds it can use it. Steal one from a log file, a browser’s storage, a proxy, or a compromised laptop and you are the user, until it expires. DPoP—Demonstrating Proof of Possession at the Application Layer, standardised as RFC 9449 in September 2023—removes that property by tying the token to a key that only the legitimate client holds.

The Mechanism

The client generates a key pair, usually ephemeral and held in memory or in non-extractable browser storage. With every request it sends a short-lived JSON Web Token called a DPoP proof in a header of the same name, signed with the private key, containing the public key in its header, the HTTP method and URI of the request, a timestamp, a unique identifier, and a hash of the access token.

POST /payments HTTP/1.1
Host: api.example.co.uk
Authorization: DPoP eyJhbGciOiJFUzI1NiIsInR5cCI6ImF0K2p3dCJ9...
DPoP: eyJ0eXAiOiJkcG9wK2p3dCIsImFsZyI6IkVTMjU2Iiwiandr...

When the token is first issued, the authorisation server records a thumbprint of that public key inside the access token, in the confirmation claim cnf with member jkt. The resource server then checks three things: the access token is valid, the proof is correctly signed by the key whose thumbprint is in the token, and the method, URI, timestamp, and identifier in the proof match this specific request.

An attacker who steals the access token alone has nothing. They cannot produce a proof without the private key, and they cannot reuse a captured proof because it is scoped to one method, one URI, and one moment.

Why It Appeared

Token theft overtook password theft as the practical route into cloud accounts. Adversary-in-the-middle phishing kits capture the post-authentication session rather than the credential, which defeats most second factors; infostealer malware harvests tokens from browser storage by the million; and tokens leak through logging, referrer headers, and misconfigured proxies.

None of that is fixed by stronger authentication, because the authentication already succeeded. Sender-constrained tokens address the problem at the other end: make the token useless to anyone but the client it was issued to. The same reasoning drives token protection features in conditional access and the device-bound session proposals now appearing in browsers.

DPoP Compared With Mutual TLS

DPoP (RFC 9449)mTLS binding (RFC 8705)
LayerApplicationTransport
Works through TLS-terminating proxiesYesOnly with header propagation
Suits public clients and browsersYesRarely
Certificate lifecycle neededNoYes
Per-request costSignature on every callHandshake only

Mutual TLS binding is stronger where you control both ends and the network path, which is why open banking and other regulated API programmes adopted it first. DPoP is the pragmatic choice for single-page applications, mobile apps, and any architecture with load balancers or API gateways terminating TLS. Both are far better than an unconstrained bearer token.

What DPoP Does Not Do

It protects the token, not the client. If malware is running inside the process that holds the private key, it can sign proofs too, and the binding provides nothing. DPoP is a defence against token exfiltration and replay, not against endpoint compromise.

It also demands correct validation at the resource server, and partial implementations are common. A server that accepts the access token and ignores the proof header has gained nothing while believing otherwise. Check that your implementation verifies the thumbprint against the confirmation claim, enforces a tight clock skew window, rejects reused proof identifiers within that window, and validates the method and URI rather than only the signature. Nonce support, where the server issues a value the client must include, raises the bar further against pre-computed proofs.

Deployment is incremental. Many authorisation servers support DPoP alongside bearer tokens, so you can enable it for the highest-value APIs—payments, administrative operations, bulk data export—before extending it. Keep revocation working regardless, because binding reduces the value of a stolen token but does not replace the ability to kill a session.

At Privia Security, we test token binding and API authorisation logic through our web application security services.

FAQ – Frequently Asked Questions

Does DPoP Replace Bearer Tokens?

It changes how they are presented rather than replacing the concept. The access token is still issued by the authorisation server, but it carries a thumbprint of the client’s public key and is sent with a signed proof on every request. Without that proof the token is rejected, so possession of the token alone no longer grants access.

Is DPoP Better Than Mutual TLS for Token Binding?

Neither is universally better. Mutual TLS binding is stronger and cheaper per request, but it needs a client certificate lifecycle and breaks where a proxy terminates TLS without propagating certificate details. DPoP works at the application layer, so it survives gateways and suits browser and mobile clients that cannot hold certificates. Many organisations use mTLS internally and DPoP for public clients.

Does DPoP Protect Against Malware on the Device?

No. If an attacker executes code inside the client, they can ask the same key to sign proofs for them. DPoP defends against token exfiltration and replay from logs, proxies, browser storage, or an intercepted session. Endpoint compromise needs endpoint controls; binding simply means the stolen token cannot be used from somewhere else.

What Do Resource Servers Get Wrong When Implementing DPoP?

The common failure is validating the signature but not the binding. A correct implementation checks that the proof’s public key thumbprint matches the confirmation claim in the access token, enforces a narrow timestamp window, rejects replayed proof identifiers within that window, and verifies that the method and URI in the proof match the actual request. Skipping any of those leaves replay possible.

Related Terms

You May Be Interested In These