Update review · October 26, 2026
Authorization Posture Mechanism (APM) Update Summary
What Changed Since -02
There's one big change. The draft now includes a patent licensing statement.
The rest of the changes are cosmetic: the date, the author's email and postal address, a now-correct header for Appendix B, renumbered References, and a reference bump to attestation-based client auth -11. The protocol text, pseudocode, wire format and security considerations are word for word the same as -02.
The Patent Disclosure
The new "IPR Considerations" section says Sanctum SecOps LLC has pending US patent applications (64/080,137 and 19/698,870) that cover APM. Brian Vicente is named as the inventor.
The licensing terms are RAND, and they "may include reasonable fees or royalties." A license requires a separate written agreement. It covers only the compliant portion of a product. It explicitly excludes optional features, hosted control planes and managed services. It also includes reciprocity and defensive suspension clauses.
This matters more than anything else in this revision. The OAuth WG has historically built on royalty-free terms. A RAND-with-fees license on a mechanism that wraps RFC 8705, RFC 9449 and RFC 9396 is going to be a hard sell for adoption. The exclusion of "hosted control planes" and "managed services" is a particular problem, since most enterprise authorization servers (Okta, Entra, Ping) run as exactly that.
Licensing terms normally belong in a Datatracker IPR disclosure, not in the draft body. The draft acknowledges this by saying the Datatracker disclosures "remain authoritative." So why is it in the document at all?
Was Previous Feedback Addressed?
No. None of the substantive points from the -02 review were addressed.
- Decision logic still implementation-defined.
ComputePostureDegradationandDispatchGraduatedOutcomeare still out of scope. Two conforming implementations can still return different outcomes for the same inputs. - "Deterministic" claim still in the abstract. REQ-3 still requires that identical inputs produce identical outcomes within one implementation. That isn't interoperability.
- Posture signal format still not mandated. The baseline JWT shape is still only a recommendation.
attestation_classvalues and their ordering are still implementation-defined. reason_codestill opaque. Clients still can't use it to take action. The example in Appendix A usesDEVICE_COMPLIANCE_CHANGED, which isn't in the list of acceptable codes in Section 5.1.- NIST ZTA mapping still present. It's still in Appendix B.
- Implementation status still points to a private repo. The repo is "private until counsel hand-off." An implementation nobody can look at doesn't help anyone test interoperability.
The draft also still defers to "the Applicability section of the corresponding Sanctum SecOps publication" for tuning parameters. That publication isn't cited in the references. Readers can't see the part that makes the mechanism actually work.
New or Removed Mechanisms
None. The protocol is unchanged.
So What Does This Mean?
The technical gaps from -02 are still open. On top of that, implementers now face a possible royalty for whatever parts are specified. The specified surface is mostly a new authorization_details type plus an HTTP status mapping, so the cost-to-value ratio for implementers got worse.
If the author wants WG adoption, the most useful next steps are:
- Offer royalty-free licensing terms, or say clearly why RAND-with-fees is necessary.
- Narrow the scope to the outcome wire format and HTTP binding, or specify the decision function normatively.
- Make the reference implementation and the "Applicability" document public.
Until then, treat this one as a vendor proposal rather than a candidate WG work item.