SCIM IETF 124

Chairs: A. Parecki, N Cam-Winget
Scribes: P. Dingle, D. Zollner

Meeting Start / Chair Comments

• Reminder of new notewell
• Update on current wG document statuses
• Cursor-based pagination is now an RFC - RFC 9865 (yay!)
• SCIM Device Model and SCIM Profile for Security Event Tokens are in
  RFC Editor Queue

Roles and Entitlements (Unmesh) - 10min

https://datatracker.ietf.org/doc/draft-ietf-scim-roles-entitlements/

• Originally published by Danny and adopted by WG, Okta is now
  proposing changes
• minimal changes - changed some attributes to optional from required
• included resource schema
• Asking for feedback on two new URNS (see deck)
• Core SCIM Standard only allows SCIM clients to set "roles" and
  "entitlements" but does not provide discovery mechanism
• Slides covering new proposed schema for role and entitlement
  resources
• New RolesAndEntitlements attribute proposed in draft, links role +
  principal(user, non-human identity, etc) + application (Check if
  this is accurate - may be wrong / tied to later slides from Prithvi)
• Request for feedback on new draft, open for questions
• Eliot Lear asked about applicability for other resources besides
  Users, such as devices
• Pam Dingle asked about existing format of "roles" and "entitlements"
  attributes on the core User schema in SCIM (RFC 7643) and the
  complexity(?) of the complex attribute value

• Danny: Agree with Pam's comment. Some of the complex objects are set
  up with value, display, type, and primary -- across the various SCIM
  service provider implementations that have implemented SCIM "roles"
  on Users, the value sub-attribute is the only thing consistently
  used. type and primary are rarely used, major clients do not
  consistently interpret and many have wildly different
  interpretations.

  • Also - have seen a couple of drafts published for AI agents etc
    that include roles/entitlements, mentioning that they correspond
    to user roles/entitlements. Suggest concern that perhaps other
    use cases

• Nancy suggest taking concerns either to a PR or the mailing list

SCIM Agents and Agentic Applications (Macy) - 15min

https://datatracker.ietf.org/doc/draft-abbey-scim-agent-extension/

• Why SCIM? History is repeating itself. SCIM for users was invented
  to lower the burden of implementers. Similar situation has arisen
  for agents. All seem to have component of cross-domain identity
  management
• Goal: keep it simple, why not have a new identity for SCIM. eg:
  discover agents implementing an A2A agent through scim interface
• Proposing new resource + associated core schema. Agents are
  workloads with unique identifiers and privileges. Should have its
  own type of identity.

• Feedback from the mailing list - why not propose another schema
  implemented in addition to user.

  • User stuff really doesn't apply to agents, eg password,
    timezone, etc

• What is an agent? (Fill out notes after reviewing AI
  notes/recording..)

• Slides detailing schema for new resource type + reasoning
• Goal: exposing discrete clients for purposes of discovery.
• Open to feedback and

Elliot: like this approach, but document is loose, some of the semantic
values are a bit loose such as agent type
X.509 attributes: should be clear as to how those certificates are meant
to be used. Is the full cert needed?
Subject - could be clarified. Looks like pretty good beginning
Pieter K: great to see things like this appearing. You define an agent
as a workload with unpredictable behavior. Surprised to see much of the
nature you describe attributed to a user. Could you start with a
workload rather than starting with a user?
Danny: Worth considering whether this draft needs to only be focused on
agentic identities or zoom out to workload identities, and agents could
be a subset.
Max Gerber: echoing last 2, can see use cases for examples without an
LLM in the mix, like the proposal.

Agentic Identity Schema (Mark) - 15min

https://datatracker.ietf.org/doc/draft-wahl-scim-agent-schema/

• Pam presenting on behalf of Mark Wahl. Mark / Pam interested in
  collaborating with Macy

• How will large enterprises manage agents

  • Pre-existing relationships between identities already exist and
    are STRONG
  • Portability of identity data (agents) between systems (identity
    providers, etc.)
  • Ability to update/delete is important
  • Need to be able to attach identity governance workflows to
    agent/non-human identities

• SCIM has been used to populate(push) data (dashboards, application
  identity stores), but now due to AI will be more push + pull to
  constantly review and adjust data between systems

• Slide detailing proposed schema for AgenticIdentity resource

  • Focus on proposed oAuthClientIdentifiers attribute, showing the
    binding of the agentic identity to a domain

• Slide comparing/contrasting registration standards (OAuth Dynamic
  Client Registration, RFC 7591 + 7592), Client ID metadata Document
  (CIMD), and SCIM

  • SCIM fits well for what is needed because of support for create,
    update, delete, and read
  • Support needed for migrating versions

• Question from George Fletcher

  • Multiple problems that need to be addressed, discovery, etc.
  • Enterprises are more structured (? check recording)
  • Authorization side - what is the agent authorized to do. SCIM is
    powerful here, some other standards have problems
  • May be a need for instance identifiers for LLM agents
  • What are we trying to accomplish with SCIM and what are we not
    trying to accomplish
  • Pam response:
    • Latency issue between registration & requesting access
    • One major concern is unauthenticated smashing of
      registration endpoints

• Question from Roberta Robert

  • With MCP group
  • Two questions
    • Issue with tooling / tooling registry
    • Extensions of ASIM(?) or SCIM(?)
    • When links are altered..
    • Multi-hop federation (agent calls tool A, which calls tool
      B, etc..) - is this covered in this

• Justin Richer comment

  • Draft seems to conflate agents and OAuth clients
  • Pam response: This draft doesn't attempt to set schema, assumes
    it will exist elsewhere and have to be adopted/used

• Peter Liu question

  • How often would the record/identity be updated?
  • Pam response: This defines protocol/schema, that is
    implementation specific (Pam confirm?)

Role Assignment (Prithvi) - 10min

https://datatracker.ietf.org/doc/draft-poreddy-scim-role-assignment/

• Draft proposes RoleAssignment resource to represent role assignments
  to resources
• User roles and entitlements are global, no link between role and
  resource context
• Workarounds break at scale - e.g.: embedding resource info in role
  name
• We need a standardized way to handle this

• Draft introduces new RoleAssignment resource, captures who + what
  role + where

  • What principal has access
  • What access
  • To what resource

• Enables discoverable, scoped role bindings

• Prior work

  • Cmomplements roles/entitlements draft
  • that draft defines catalog resources (/roles, /entitlements)
  • RoleAssignment defines bindings (who has which role in which
    context)
  • Together: Discovery -> Definition -> Assignment

• This draft is complementary and is applicable and implementable in
  existing enterprise systems

• design choices: standalone resource rather than - extending user
  roles. This means you can query/manage without touching user records
  at scale.

• If you need to change scope you create new objects
• Status is always computed, means the same thing across all
  assignments.
• Next: feedback on 3 areas: subject scope role relationship, larger
  deployments
  Max Gerber: How can a client discover what scopes are applicable to
  what roles?
  Prithvi: you can query them to see what is available, that is built
  into the spec, we are using what is already there.

Nancy: I think it is within the charter
Deb Cooley (AD): I think it is time for a new charter discussion. Things
we said we would do first have not happened.
Nancy: Use cases did start talking about the SaaS aspect from a user
perspective
Deb: I think you have interest, I got questions about the device draft
and there was a path to agreeing, but it is time to update the base
specs:
Nancy: the roles spec is intended to be one of those updates. We can
take to the mailing list but I'm not seeing... if you want to progress
these specs you need to update the base specs too. Let's continue on the
mailing list