Uber Engineering has outlined a framework for deciding whether an AI agent’s proposed action fits the user’s request, the agent’s purpose, its permissions and the interaction’s context. Uber says ordinary identity and API controls can check access, but not whether an action is appropriate.
The framework covers three stages: registration and security checks before deployment, runtime checks by AI Guards and Agent Guard hooks, and ongoing review using execution traces and replayable interactions. Uber says explicit, agent-specific intents and constraints were its strongest control, though setting them up takes work that grows with the number of workflows.
Why it matters
For teams building AI agents, Uber’s approach offers a way to check actions in context, but its per-agent rules may require substantial setup; the account does not explain how to enforce them on vendor agents.