Uber Enhances AI Agent Identity and Access Control Framework

In 2025, Uber is implementing updates to its identity and access technology stack to enhance the accountability of AI agents. As the company embraces the growing role of AI in its operations, these updates aim to tackle issues related to the traceability and transparency of actions performed by autonomous agents.

Context of the Updates

Uber's internal Agent platform, launched in early 2025, enables teams to develop and deploy AI solutions at scale. This platform is supported by the Model Context Protocol (MCP), which integrates with Uber's extensive microservices infrastructure. However, as the use of AI agents grows, the need for rigorous oversight has become evident. The challenge is ensuring that every action taken by an agent can be traced back to a specific human user, preserving a clear line of accountability.

Traditional identity models, primarily designed for human users and workloads, do not adequately capture the complexities of AI agent interactions. Often, as agents execute tasks, the original context of their actions is lost, complicating auditing and compliance efforts.

Identifying the Core Problems

Uber has pinpointed two major issues with current identity models:

1. Inadequate Representation of Agency: Existing frameworks fail to effectively represent the agents' ability to act on behalf of human users. This creates a disconnect between the actions performed and the identities responsible for them.
2. Loss of Provenance: As agents interact with various systems, the context of their actions is frequently lost, complicating audits and incident responses. This lack of continuity can hinder the enforcement of security protocols and policies.

Addressing these issues is essential as Uber seeks to use autonomous agents more effectively while ensuring compliance and security in its operations.

Architectural Enhancements

To address these challenges, Uber is extending its Zero Trust Architecture specifically for AI agents. Key components of this enhanced architecture include:

• Agent Registry: Serves as a source of truth for agent registration, linking AI agents to their respective workloads managed through Kubernetes.
• AI Agent Mesh: Enables communication between agents, allowing them to collaborate on tasks. This mesh is supported by short-lived JSON Web Tokens (JWT) for secure authentication.
• Security Token Service (STS): Issues scoped tokens for each agent interaction, ensuring that only authorized agents can perform specific actions.
• MCP Gateway: A central system that mediates calls from the AI Agent Mesh to Uber’s internal systems, enforcing policy compliance and security measures.

These components work together to provide a framework that maintains security while allowing for high developer velocity, crucial for Uber’s fast-paced environment.

Implementing Agent Identities

A significant aspect of the updates is the provision of verifiable identities for each AI agent. By integrating a cryptographic identity framework, Uber ensures that every agent can authenticate itself and maintain a unique identity throughout its operational lifecycle. This is achieved through a workflow where each agent requests a JWT from the STS, which includes a complete history of the actor chain involved in its actions.

This process allows for comprehensive audit logs, facilitating greater visibility and trust in the actions carried out by AI agents. Uber's approach ensures that every operation can be traced back to the original user, addressing both security and compliance requirements.

The Road Ahead

Looking ahead to 2026, Uber plans to continue enhancing its identity and access management systems to support the growing use of AI agents. By focusing on dynamic access control and unified policy enforcement, the company aims to create a cohesive architecture where identity, risk, and policy work together effectively.

As the adoption of AI agents increases, maintaining a secure, auditable, and efficient operational framework will be essential. Uber's proactive measures reflect its commitment to advancing technology while ensuring security and accountability, paving the way for the future of AI in enterprise environments.