IndyKite AgentControl now enforces least privilege for every request, carried across every agent in a workflow and grounded in the enterprise's own data.
November 17, 2026 – IndyKite today announced Intent-Based Access Control (IBAC) in IndyKite AgentControl. IBAC turns what a subject asked for into a structured intent, carries it across every agent and tool in a multi-agent workflow, and enforces it at every step. Agents get exactly the access their task needs, and nothing more, while IndyKite's knowledge graph decides what data each step may touch based on relationships and trust.
Enterprises moving multi-agent AI into production face a gap in how agents are controlled today. On-behalf-of delegation, used by most identity platforms, gives an agent its own identity and records who it acts for. But it also hands that agent the full breadth of the person's permissions, and passes them down every hop of the workflow. Multi-agent workflows are where permissions leak.
IndyKite IBAC builds on the delegation model that identity providers organizations already run, and scopes each agent's authority to the single task the subject asked for. It is designed around three capabilities:
- Least privilege for every request. AI is used once, to translate the subject's request into a structured intent using OpenID AuthZEN semantics. From that moment, every decision is made by rules rather than a model's judgment. The same request gets the same answer every time, and every decision can be proven in an audit.
- Carried across every agent. The intent is sealed in a signed IndyKite token attached to every actor in the delegation chain. It travels agent to agent and agent to tool, across A2A and the Model Context Protocol (MCP), whichever identity provider issued the identity. Each downstream agent's intent can only match or narrow the original, so privileges cannot escalate.
- Grounded in your data. Intent is expressed in the entities and relationships of the organization's own knowledge graph. Access to the data itself is then decided on relationships, consent, sensitivity, provenance and trust.
“When the limits of a task hold across the entire workflow, the scope of what can be automated dramatically expands, while risk is contained,” said Lasse Andresen, Founder and CEO of IndyKite. “With IBAC, enterprises can now get enforcement they can predict, repeat and prove, carried across every agent and grounded in how their business and data actually relate.This level of control is transformative, especially for enterprises working in regulated industries.”
IBAC delivers:
- A contained blast radius. If an agent is compromised or drifts, it stays bounded by the original intent.
- Decisions regulators can trust. Rules-based enforcement gives regulated industries outcomes they can defend to auditors.
- End-to-end traceability. Every decision is recorded with the full delegation chain, from each database query back to the original request, without instrumenting agents.
- No rebuilds. The Agent Gateway sits in front of existing agents and MCP servers as a proxy or side-car, built on OAuth2, OpenID AuthZEN, A2A and MCP.
IBAC is the first step in IndyKite’s broader investment in governing agentic AI. Today, IBAC scopes access to the task a subject asked for. IndyKite is now extending the same model to the agents themselves, so that an agent’s declared purpose becomes a boundary it cannot cross, whoever or whatever invokes it. As agents take on more of the work people once did, IndyKite’s aim is that no agent acts beyond the purpose it was given.
Intent-based Access Control is available today with IndyKite AgentControl. The capture layer, the IndyKite Intent Agent, is available as open source at github.com/indykite/intent-agent. Organizations can try IndyKite or book a demo at IndyKite.ai







