Identity and access management architecture

Identity and access management architecture

Identity and access management architecture at design depth (authentication, authorisation, federation, lifecycle, governance), applied to AI systems and agents

• Architected and productised identity services (reference architecture, platform evaluation and selection, service definition), not implementation only
• AI agent identity lifecycle: registration with a human sponsor, declared scope, short-lived credential issuance, revocation; hands-on with at least one of Microsoft Entra Agent ID, Okta cross-app access, SailPoint Agent Identity Security, Cisco/Astrix
• AI access governance: bringing models, agents and datasets under the same entitlement model and access certification as applications, through SailPoint ISC or Saviynt at registry and API level, and SCIM 2.0
• Delegated authorisation at protocol depth: how a user's authority becomes an agent's authority via OAuth 2.1, token exchange (RFC 8693) and on-behalf-of flows, sender-constrained tokens (DPoP, mTLS), OIDC claims and audiences
• Built or integrated an authorisation system in production (policy engine, authorisation service or token-level IdP integration)
• Policy as code for agent scopes and tool permissions: OPA/Rego or Cedar
• Model Context Protocol and agent frameworks: where identity is asserted in a tool call, tool allow-listing, per-task isolation
• Prompt-time entitlement enforcement: RAG architecture with retrieval scoped to the asker's identity entitlements (permission filtering at retrieval, index scoping per asker)
• AI gateway / prompt control layer deployment (Microsoft Purview, Zenity, Lasso or open-source) integrated with the identity provider for the asker's token and the entitlement model for their rights
• Microsoft Entra ID or Okta at token issuance, conditional access and session level, where an agent's delegated authority originates and is revoked
• Non-human identity governance and SPIFFE/SPIRE workload identity as the source of agent credentials

Good to have

• Microsoft, Google, AWS and Anthropic agent and model platforms; sovereign model hosting (Mistral, Aleph Alpha)
• Telemetry modelling in an open schema (OCSF)
• EU AI Act human-oversight obligations; GDPR and works-council constraints on prompt inspection
• Awareness of AI security and MITRE ATLAS (prompt injection, model safety), which is adjacent to this role rather than part of it

Candidates will rarely tick every must-have, and that is expected. The ones that should not be waived are:

• the service-architecture experience on all three
• SOC fluency for ITE
• key ceremony and PKI design depth for Machine Identity, and
• the protocol-level authorisation experience and IAM architecture depth for Identity for AI

Work Location: Bangalore & Mumbai

Here at Atos, diversity and inclusion are embedded in our DNA. Read more about our commitment to a fair work environment for all.

Atos is a recognized leader in its industry across Environment, Social and Governance (ESG) criteria. Find out more on our CSR commitment.

Choose your future. Choose Atos.