True Delegation — On-Behalf-Of Flows

The most dangerous anti-pattern in AI deployments today is allowing an agent to impersonate a human user via their own token. When an agent acts under a user's identity, downstream systems cannot distinguish between the human and the machine. Indigo helps you establish accountability and make EU AI Act compliance structurally possible.

• Implement OAuth 2.1 On-Behalf-Of (OBO) flows per RFC 8693 Token Exchange so every access token carries two distinct identifiers: the human principal who delegated authority, and the specific agent instance executing the task.
• Every action is definitively linked to both, creating a non-repudiable audit trail that satisfies regulatory requirements and security forensics.

Forcing Token Exchange ensures both delegation and appropriate authorization.

Centralized Authorization & MCP Security

Distributed authorization: where each agent or service makes its own access decisions; creates a patchwork of inconsistent policies that attackers can exploit. Without a central control plane, Shadow AI can proliferate. We help you see when employees connect ungoverned agents to sensitive corporate data.

• Decouple the Policy Decision Point (PDP) from the Policy Enforcement Point (PEP), centralizing guardrails across your entire agent estate.
• Disable open Dynamic Client Registration to block anonymous shadow agents, and enforce Enterprise Managed Authorization at the gateway layer; Including guardrails that can enforce PII masking before data reaches an LLM, and rate limits and token budgets to contain anomalous behavior.

Automated Agent Lifecycle Management

Enterprises operating hundreds of concurrent AI agents cannot rely on manual credential rotation or ad hoc offboarding. An agent that is never de-provisioned—because no one tracked its creation—is an open door. We ensure your organization isn’t accumulating debt.

• Leverage robust IAM provisioning tools and open protocols like SCIM 2.0 to automate agent onboarding and offboarding with the same rigor applied to human employees.
• Purpose-built platforms like Amazon Bedrock AgentCore (Include Other Hyper Scaler?) provide agent lifecycle capabilities out of the box, which we integrate into your existing IAM governance layer.
• Actualize a single operation that instantly de-provisions a compromised or retired agent and revokes its access across every integrated system simultaneously.
• Discover agents that are operating within your infrastructure and onboard them into your governance platform.

Meaningful Human Oversight & Recursive Delegation

Full autonomy is appropriate for low-risk, reversible operations. For high-stakes actions—deleting data, initiating financial transactions, modifying access policies—autonomous execution must pause and seek human approval. And when agents orchestrate sub-agents, the chain of authority must be properly bounded at every hop. We ensure that there is a “human-in-the loop.”

• Implement Client Initiated Backchannel Authentication (CIBA) to push out-of-band approval requests directly to a user's trusted device before any high-risk action executes.
• For multi-agent architectures, we enforce Scope Attenuation: downstream sub-agents can never be granted permissions that exceed those of the orchestrating agent, preventing privilege escalation through the delegation chain.