P0 Security extends its authorization control plane to workloads and AI agents
General availability for non-human identity lifecycle management plus runtime authorization for agents, with one enforcement model that combines the invoking user, the agent, the tool, and the target resource.
P0 Security has made its non-human identity lifecycle management and authorization control plane for agents generally available, extending the platform from human users to service accounts, workloads, and first-party AI agents. The agent control plane evaluates access when the agent acts, combining four inputs into one decision: the invoking identity, the agent identity, the tool authorisation, and the resource entitlements. Standing permissions are replaced with ephemeral least-privileged grants enforced at runtime. Each identity type gets its own workflow: humans request temporary privileged access through approvals, workloads request programmatically through platform integrations, and agents are authorised during tool execution. Integrations span AWS Bedrock, Google Vertex AI, and Microsoft's agent platform.
Why it matters
The four-input model is the useful detail. Most agent authorisation collapses to one question, what the agent is allowed to do, and that is wrong. An agent invoked by a support engineer and the same agent invoked by a finance analyst should not resolve to the same permissions, and an agent acting on a schedule with no human behind it is a third case again.
Carrying the invoking identity through to the resource decision is least privilege applied honestly to delegation, and it is what separates real authorization from a token with a broad scope.
The scoping word is first-party. This governs agents you build and deploy, not vendor-hosted assistants acting on your data, which remain a contractual problem rather than a technical one. Compare approaches in our authorization tools ranking.
Source: Business Wire