Uber's Secure AI Framework Governs Autonomous AI Agents Across Their Lifecycle

Uber Engineering has unveiled a Secure AI Framework designed to govern the lifecycle of autonomous AI agents at Uber. This framework addresses the limitations of traditional identity and API controls in securing AI agents, which can dynamically reason over external content and take real-world actions. Uber's framework is composed of three layers: pre-deployment, runtime enforcement, and post-deployment continuous assurance.

During provisioning, agents are registered in the Uber AI Registry, which establishes identity, ownership, persona, permitted intents, constraints, dependencies, and risk classification. This process is automated using Argus, which analyzes the agent's source code, prompts, configurations, dependencies, exposed endpoints, and tool-invocation paths to identify security weaknesses.

At runtime, the framework incorporates behavioral authorization, evaluating whether an agent's actions are appropriate in a given context through three components: the AI Registry, AI guards, and Agent Guard runtime hooks. After deployment, the focus shifts to continuous assurance, which relies on end-to-end observability, auditing, explainability, and replay.

Uber has identified several challenges in implementing security controls in production, such as balancing policy restrictiveness, minimizing runtime latency, reducing developer friction, and moving towards agent-specific policies for more contextual and precise enforcement.

Brief written by urgent.news from InfoQ's own syndicated text. Machine-written — may contain errors; check the original before relying on it.