sources · source-idac-390-agentic-ai-identity
#390 - Identity Management for Agentic AI with Tobin South
Tobin South explains why AI agents strain identity systems designed for deterministic users and applications. The discussion separates impersonation from delegated authority, argues that agents need identities and credentials distinct from the people who invoke them, and connects governance to bounded scopes, traceable delegation chains, oversight, and responsibility for actions. South treats standards work as necessary infrastructure for both consumer assistants and enterprise deployments, while also describing Model Context Protocol as an interoperability layer whose security properties still depend on authentication and authorization. The episode's practical emphasis is that identity teams should prepare policies and architectures before agent adoption becomes widespread, without assuming current protocols or shared user credentials are adequate.
Open the canonical original source
Source at a glance
- Source type
- podcast episode
- Publisher
- IDAC Corp. / Identity at the Center
- Source or access date
- 2025-12-08
- Analysis depth
- deep
- Rights treatment
- link and paraphrase
- Human review
- Murray Newlands · 2026-08-16
Important limitations
- Expert perspective rather than empirical study
- Recording date not publicly established
- No official transcript
- Host banter and event promotion occupy the opening and closing sections
Source-located statements
10 reviewed statements are indexed from this source.
- South says delegated authority becomes materially harder with AI agents because nondeterministic systems can act externally, creating a need for a dedicated standards community to define safer identity infrastructure. (00:09:11)
- South distinguishes an assistive AI that remains close to a user's direct interaction from an agent exercising delegated authority after receiving only a high-level goal, and from an autonomous service-like agent. (00:12:53)
- South recommends giving an assistant selective service access and preserving the ability to distinguish actions taken by the assistant from actions taken by the principal, instead of sharing the principal's full credentials. (00:18:28)
- South assigns different governance responsibilities to consumer-assistant builders, web and standards bodies, and enterprises, with enterprise deployments requiring explicit identity and access-management policies and procedures. (00:19:15)
- South rejects a single universal fix and recommends adapting existing web identity, delegated-authority, and enterprise identity systems so agents can be declared, linked to a human principal, provisioned, and deprovisioned. (00:21:50)
- South predicts that access permissions, human oversight, and service-access management—not only model capability—will constrain the usefulness of long-running agents. (00:24:51)
- South warns that responsibility for consequential agent actions is legally and operationally unclear, and argues for guardrails that combine task alignment, bounded access, and explicit analysis of business-process risk and liability. (00:27:37)
- South recommends agent-specific interfaces that restrict available actions and preserve an agent identity, rather than allowing an agent to use a human-oriented command interface that makes its activity appear to be the user's own. (00:30:16)
- South describes Model Context Protocol as an agent-oriented wrapper around APIs that translates a natural-language interaction into access to tools while relying on familiar concepts such as scopes, API keys, and OAuth consent. (00:32:32)
- South explains that recursive delegation moves the human principal farther from each downstream action, so every sub-agent should receive only the narrower permissions needed for its task rather than inheriting the principal agent's full scope. (00:42:44)
Evidence lineage and transparency
| Relationship field | Linked identifiers |
|---|---|
| speaker ids | person-jeff-steadman, person-jim-mcdonald, person-BATCH-2026-005-tobin-south |
| organization references | orgref-idac, orgref-openid-foundation |
Machine review: ready for human review. Human review: approved. Workflow: published.
Complete structured record
- source id
- source-idac-390-agentic-ai-identity
- canonical title
- #390 - Identity Management for Agentic AI with Tobin South
- alternate titles
- Identity at the Center #390: Identity Management for Agentic AI with Tobin South
- source type
- podcast_episode
- series or parent source
- Identity at the Center, episode 390
- publisher
- IDAC Corp. / Identity at the Center
- channel
- Identity at the Center
- speaker ids
- person-jeff-steadman, person-jim-mcdonald, person-BATCH-2026-005-tobin-south
- author ids
- institutional author
- Unknown
- organization references
- orgref-idac, orgref-openid-foundation
- recorded at
- Unknown
- event date
- Unknown
- published at
- 2025-12-08
- updated at
- Unknown
- duration seconds
- 3353
- language
- lang-en
- translated title
- Unknown
- translation method
- Unknown
- geography of speaker
- geography of organization
- geo-global
- geography discussed
- geo-global
- study geography
- original url
- https://anchor.fm/s/c5fefcc/podcast/play/111687080/https%3A%2F%2Fd3ctxlq1ktw2nl.cloudfront.net%2Fstaging%2F2025-10-25%2F01f889a3-6ebf-b82d-8d03-61c6114f10df.mp3
- canonical url
- https://podcasts.apple.com/us/podcast/390-identity-management-for-agentic-ai-with-tobin-south/id1471899975?i=1000740200992
- archived url
- Unknown
- embed url
- https://open.spotify.com/embed/episode/6djrGPlNiIlJZmBpcsR2KY
- doi
- Unknown
- canonical identity status
- verified_original_episode_across_rss_apple_spotify_and_openid_media_index
- repost status
- original_episode_with_listening_platform_mirrors
- original source id
- Unknown
- rights status
- link_and_paraphrase
- ownership status
- third_party
- relationship to off
- none
- transcript status
- unofficial_research_aids_only
- transcript source
- Podscan public preview plus local machine transcription checked against the original audio
- transcript republication permission
- not_documented_do_not_publish_full_transcript
- chapter markers
- [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object], [object Object]
- analysis basis
- Complete original audio reviewed with official chapter markers, platform metadata, an unofficial transcript preview, and a locally generated timestamped research transcript; statements were checked against the audio dialogue and attributed only where the guest response was clear.
- topics
- topic-ai-agents, topic-ai-governance, topic-cybersecurity, topic-non-human-identity, topic-ai-liability, topic-enterprise-infrastructure
- executive roles
- role-ciso, role-cio, role-cto, role-general-counsel, role-board-director
- original abstract
- Tobin South explains why AI agents strain identity systems designed for deterministic users and applications. The discussion separates impersonation from delegated authority, argues that agents need identities and credentials distinct from the people who invoke them, and connects governance to bounded scopes, traceable delegation chains, oversight, and responsibility for actions. South treats standards work as necessary infrastructure for both consumer assistants and enterprise deployments, while also describing Model Context Protocol as an interoperability layer whose security properties still depend on authentication and authorization. The episode's practical emphasis is that identity teams should prepare policies and architectures before agent adoption becomes widespread, without assuming current protocols or shared user credentials are adequate.
- inclusion rationale
- This is a complete, long-form interview with the lead author of the OpenID agentic-identity report and directly addresses delegation, credentials, governance, liability, MCP, and recursive authority.
- source quality dimensions
- {"primary_spoken_source":"high","identity_stability":"high","access_completeness":"complete","independence":"interview_not_peer_reviewed"}
- methodology quality
- {"study_design":"expert_interview","empirical_method":"none","attribution_method":"dialogue_context_and_audio_verification"}
- study design
- expert_interview
- sample
- {}
- population
- Unknown
- date range
- {}
- funding
- IDAC Corp. states that the regular weekly podcast does not sell advertising or use commercial sponsorship contracts; no episode-specific funding was identified.
- sponsor
- Unknown
- peer review status
- not_applicable
- findings
- limitations
- Expert perspective rather than empirical study, Recording date not publicly established, No official transcript, Host banter and event promotion occupy the opening and closing sections
- correction ids
- retraction status
- Unknown
- content hash
- sha256:779b09ce66832cb86acf34c091302cc3a653f5487dd4b640bed84c16570af40d
- accessed at
- 2026-08-15
- verification status
- machine_verified_human_approved
- source depth
- deep
- publication status
- published
- workflow status
- published
- machine review status
- ready_for_human_review
- human review status
- approved
- reviewed by
- Murray Newlands
- reviewed at
- 2026-08-16T23:17:22Z
Provenance and revision history
{
"provenance": [
{
"source_url": "https://podcasts.apple.com/us/podcast/390-identity-management-for-agentic-ai-with-tobin-south/id1471899975?i=1000740200992",
"accessed_at": "2026-08-15",
"retrieval_method": "official platform metadata and original RSS audio review",
"exact_locator": "episode 390; 00:00-55:53",
"content_hash": "sha256:779b09ce66832cb86acf34c091302cc3a653f5487dd4b640bed84c16570af40d",
"batch_id": "BATCH-2026-004",
"prompt_id": "OEII-MEDIA-RESEARCH",
"prompt_version": "2.0",
"notes": "No transcript asset is included in the repository."
}
],
"revision_history": [
{
"changed_at": "2026-08-16T23:17:22Z",
"changed_by": "Murray Newlands",
"summary": "Approved for the governed-identities pilot release under the exact scope, exclusions, rights treatment, and limitations recorded in issue #18.",
"batch_id": "BATCH-2026-004"
}
]
}