Review · August 26, 2026
SCIM 2.0 IPSIE Profile
draft-schreiber-scim-ipsie-profile-00 is an individual submission from Jen Schreiber (CrowdStrike) and Danny Zollner (Okta). It's 13 pages, Informational, and it's the SCIM half of the OpenID Foundation's IPSIE work showing up as an IETF draft.
What problem it solves
Enterprises buy a lot of SaaS apps, and every one of those apps implements SCIM slightly differently. IPSIE (the OpenID Foundation's Interoperability Profiling for Secure Identity in the Enterprise working group) is trying to fix that by defining tiered assurance levels an enterprise can put in a procurement checklist. This draft is the SCIM provisioning piece of that.
The unique thing here isn't a new mechanism. It's the packaging. The draft takes SCIM and cuts it into three cumulative Account Lifecycle levels:
- AL1 (User Deprovisioning) - the app can turn off and delete accounts when the IdP says so.
- AL2 (User and Group Management) - the IdP owns the full user and group lifecycle.
- AL3 (Role Management) - the app publishes its roles so the IdP can assign them.
AL3 is empty. It's an editor's note saying requirements come in a future revision.
How it fits with existing work
This sits on top of draft-zollner-scim-interop-profile, which is the general-purpose SCIM restriction profile. That draft narrows SCIM's optional surface (PATCH only, no PUT, eq and and filters only, no ETags, mandatory discovery endpoints). This draft adds the enterprise security requirements on top and says conforming implementations MUST also conform to that one.
So there's now a two-layer profile stack for SCIM, with a third layer implied for role management. That layering is reasonable, and the SCIM charter explicitly allows profiles: "These extensions may include profiles and schemas for interoperability in additional use cases."
Dependency count is small: RFC 7643, RFC 7644, RFC 6749, RFC 7523, RFC 8414, plus the interop profile. No new schema, no new endpoints, no new IANA registrations.
The relationship it does establish is new. SCIM core says nothing about how the SCIM client authenticates. This draft pins that down to OAuth client credentials with JWT client authentication, and it requires both sides to publish OAuth authorization server metadata. That's the first time a SCIM document has drawn a hard line from the provisioning channel into OAuth's discovery machinery.
The summary for a VP
Two parties. The Identity Service (your IdP) is always the one making requests. The Application (the SaaS vendor) is always the one receiving them. The draft is deliberate about never letting that flip.
At AL1, the IdP tells the app to deactivate a user, and the app has five minutes to kill everything: web sessions, API tokens, refresh tokens, personal access tokens, the user's SSH keys, device credentials. That list is the most valuable thing in the document. It names the access paths that survive a normal deprovisioning event and turns "we deprovisioned the user" into something you can actually audit.
At AL2, the IdP becomes the sole writer. The app has to accept user and group creates and updates over SCIM, and it has to block local account creation for IdP-managed users. Groups map to whatever roles the app has internally.
There's a hard prohibition on credentials moving over the channel. The app MUST NOT support the password attribute, MUST NOT define credential-bearing attributes in custom schemas, and the IdP MUST NOT send passwords, hashes, private keys, or TOTP seeds. This is the right call and it's stated cleanly.
At AL3 the direction reverses: the app publishes its role catalog and the IdP consumes it. Nothing is specified yet. There's an expired draft-ietf-scim-roles-entitlements from one of the same authors that presumably feeds into this.
Significant interoperability concerns
The authentication section contradicts itself. Section 4.1 says OAuth interactions MUST use client_credentials with JWT Client Authentication per RFC 7523 §2.2. The next bullet says the Identity Service MUST authenticate using HTTP Basic authentication and that transmitting client_assertion in the request body is prohibited. Those two requirements cannot both be satisfied. RFC 7523 §2.2 works by putting client_assertion in the form-encoded body. HTTP Basic is client_secret_basic, a completely different authentication method that requires a shared secret. An implementer reading this has to guess which one the authors meant, and the two guesses produce non-interoperable clients. This is the single most important thing to fix.
RFC 8414 is the wrong tool for the IdP side. The draft requires the Identity Service to publish an authorization server metadata document containing jwks_uri so the Application can validate JWT signatures. But the Identity Service in this architecture is an OAuth client, not an authorization server. Publishing AS metadata to advertise client keys is a category error. What's wanted is client metadata, either registered out of band or fetched via Client ID Metadata Document or Dynamic Client Registration. There's also no mechanism given for how either party discovers the other's metadata URL. The Application's SCIM base URL is not an OAuth issuer identifier, so there's no defined path from "here is the SCIM endpoint" to "here is the token endpoint."
The scim scope has no defined meaning. The access token MUST include the scim scope "and not grant broader permissions." Nothing says what that scope authorizes, whether it's coarse or per-endpoint, or how an AS that already has a scope namespace should reconcile. A normative wire value like this needs a definition and probably an IANA registration, otherwise every vendor ships a slightly different scim.
Nonce validation is unimplementable as written. Security Considerations says "nonces SHALL be validated to prevent replay." There is no nonce anywhere in the protocol this document describes. Client credentials has no nonce. There's no DPoP, no request signing, no challenge. Either this needs to point at a real mechanism or it needs to come out.
AL1 requires filters on attributes that AL1 never requires. Section 5.4 makes the Application support filters on externalId and on emails[value eq] and emails[type eq "work" and value eq]. But the User schema requirements including externalId live in §6.1.1, which is AL2. And emails is never a required attribute at any level. So an AL1-only implementation is told to filter on data it was never told to store. Move the schema requirements into the shared prerequisites.
The compliance statement contradicts the body on DELETE. Section 5.2 says Applications MAY allow deletion via DELETE /Users/{id}. Section 10 says an AL1 Application SHALL support DELETE /Users/{id}. Optional in one place, mandatory in the other. Given the draft also has an open editor's note about data retention and username reuse after deletion, my guess is MAY is the intent and the compliance statement is stale.
/Bulk is a MUST with no interop content. Both parties MUST support the /Bulk endpoint. The interop profile this document extends says /Bulk support is OPTIONAL, so this is a deliberate tightening. The problem is that /Bulk is one of the least consistently implemented parts of SCIM, and the draft says nothing about maxOperations, bulkId resolution, circular references, or partial failure handling, all of which RFC 7644 §3.7 leaves room to get wrong. Mandating the endpoint without constraining its hard parts adds implementation burden without buying interoperability. Either profile /Bulk properly or drop it to SHOULD and lean on multi-attribute PATCH, which the draft already requires.
The citation for /Bulk is wrong. It's cited as "[RFC7643] section 3.7." /Bulk is RFC 7644 §3.7. RFC 7643 doesn't have a §3.7 at all; section 3 stops at 3.3.
HTTP Basic tension with the layer below. The interop profile says "The use of HTTP Basic Authentication over TLS is NOT RECOMMENDED." This profile mandates HTTP Basic at the token endpoint. Different endpoints, so not a strict conflict, but a reader stacking the two documents will notice and someone will file an issue.
Mechanisms that could be removed
The performance numbers. "The Application MUST support at least 25 SCIM requests per second," per tenant for multi-tenant apps. That's a service level agreement, not a protocol requirement. It's not testable from the protocol, it will be obsolete in three years, and it has no bearing on whether two implementations can talk to each other. Same for the 5-minute propagation and revocation windows. These belong in a certification program or a procurement template, which is arguably what IPSIE actually is, but they don't belong in a normative IETF document.
"Local modifications to Users or Groups in the Application are prohibited." This is one sentence in §4.2 with no role attached, no BCP 14 keyword, and no scope. Read literally it says a SaaS app may not let an admin change a user's display name in its own UI. That's a sweeping product requirement, it's invisible on the wire, and no SCIM client can verify it. If the intent is "the IdP is authoritative for the attributes it manages," say that instead.
The internal revocation list, sort of. The requirement to kill SSH keys, PATs, and device credentials on deactivation is genuinely the most useful content in the draft, and I'd keep it. But be honest that it's an application behavior requirement rather than an interoperability requirement, and frame it as such. Nothing a SCIM client sends or receives changes based on whether the app complied.
Auditing and error-detail requirements. "SHALL be logged for audit and troubleshooting" and "SHALL NOT leak internal details" are good operational advice. Neither affects interoperability. The error format requirement is already in the interop profile.
Strip those out and this gets meaningfully shorter without losing anything an implementer needs.
Writing quality
The document is well organized and reads cleanly. Section structure maps directly to HTTP operations, which makes it easy to check an implementation against. Requirements are mostly stated as bullets with clear subjects, which is the right style for a profile. It's appropriately concise at 13 pages.
The problems are consistency, not prose. Some specifics:
- Two section headers say
/Group/{id}(singular) while their bodies correctly say/Groups/{id}. - §5.4 writes the filter attribute as
username, §6.1.1 writes it asuserName. SCIM attribute names are case-insensitive so this isn't fatal, but the interop profile has an entire section on case sensitivity and a profile should model its own rules. - §6.2.1 has a stray quote mark:
the "displayName" attribute". - The author's affiliation is printed as "Crowstrike," presumably CrowdStrike.
- §4.3 and §8 both drop into lowercase "must" inside sentences that are otherwise normative.
Three editor's notes are still in the document, one of which is an entire missing level. That's normal for a -00, but the Compliance Statement presenting a complete three-level conformance model when one level is empty overstates where the draft is.
Bigger structural point: the anchor document isn't cited. The whole framework rests on "the three IPSIE Account Lifecycle assurance levels as defined in the IPSIE Levels specification." That specification appears nowhere in the references. §4.1 also mentions the "IPSIE Session Lifecycle 1 (SL1) profile" with no citation. Those are OpenID Foundation documents. A reader who picks up this draft cold has no way to find the definitions the entire structure depends on, and the IETF has no visibility into whether those definitions are stable.
Which raises the venue question. The header says "Workgroup: IPSIE," but IPSIE is an OpenID Foundation working group, not an IETF one. Datatracker shows this as an individual draft with no stream. If the intent is for the SCIM WG to adopt this, the relationship to the OIDF work needs to be spelled out: who owns the level definitions, what happens when they change, and whether the IETF is being asked to normatively depend on a document produced elsewhere. Profiles are in the SCIM charter, so the technical fit is fine. The process fit needs a conversation.
About the authors
Jen Schreiber (CrowdStrike, [email protected]) has one published RFC, RFC 9967 from May 2026, the SCIM Profile for Security Event Tokens. Plus this draft and one expired one. Relatively new to IETF authorship but with a finished RFC in exactly this problem space, so the SCIM plus enterprise events territory is familiar ground.
Danny Zollner (Okta, [email protected]) has one published RFC, RFC 9865 from October 2025, Cursor-Based Pagination of SCIM Resources. Four active drafts right now: this one, the interop profile it depends on, draft-zollner-scim-group-members, and draft-wzdk-scim-agent-resource. Datatracker lists seven expired drafts including draft-ietf-scim-roles-entitlements, draft-sehgal-scim-delta-query, and two other individual submissions. That's a heavy current writing load and a long history of SCIM work specifically. He's clearly a central figure in what SCIM is becoming, and the interop profile plus this profile plus the group members draft are pieces of one coherent effort rather than scattered submissions.
Both authors work at IdP vendors, and it shows in the draft's shape. Every requirement is written from the perspective of an IdP pushing to an app. That's the dominant deployment pattern, so it's a defensible choice, but the SCIM charter explicitly calls out "pull vs. push based use cases" as in-scope work, and this profile forecloses pull entirely without saying so.
What I'd raise on the list
The authentication contradiction in §4.1 is the thing to fix first. It's not a nit, it's a fork in the road that produces two incompatible implementations.
After that: the RFC 8414 misuse for the client side, the undefined scim scope, the nonce requirement that has no nonce, and the AL1/AL2 layering problem where AL1 requires filters on attributes only AL2 requires. Then decide whether the performance SLAs and internal revocation requirements belong in an IETF document or in an IPSIE certification program.
The underlying idea is good and the enterprise need is real. Anyone who has run a provisioning integration knows the deactivation-doesn't-actually-revoke-anything problem this draft names, and naming it in a normative document is worth doing. The draft just needs the OAuth layer rebuilt and a clear answer on where the level definitions live.
Related reading: draft-zollner-scim-interop-profile, which this document extends and which is the more consequential of the two.