Monday, November 16, 2026 Edition No. 240
Search
Daily News Fully automated · 2 min read

Architecture AI Updates: November 16, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. Uber Describes a Three-Layer Architecture That Adds Behavioral Authorization for AI Agents on Top of Identity and Access Controls

Uber engineers described their Secure AI framework in a blog post that InfoQ covered on November 15. The design starts from the point that identity and API authorization can show whether a service may reach a resource but not whether an agent’s proposed action fits the request, the agent’s purpose and the context. It has three layers. At provisioning, agents are registered in the AI Registry, which records identity, ownership, persona, permitted intents, constraints, dependencies and risk tier, and Argus scans source code, prompts, configurations, dependencies and tool-invocation paths, including third-party MCP servers and tools. At runtime, the Registry acts as the policy administration point, AI Guard is the decision point (small specialized models hosted locally and run in parallel to meet latency budgets), and Agent Guard hooks at agent and tool boundaries enforce each verdict before an action proceeds. After deployment, a unified execution trace links agent actions, tool calls and security decisions, and recorded interactions can be replayed against candidate guardrails. Uber lists the production challenges as balancing policy strictness against false positives, keeping latency low and reducing developer friction; neither the blog nor InfoQ gives measured latency or false-positive figures. InfoQ notes commenters’ concern that per-agent personas and constraints scale in cost with the number of workflows, and an open question about vendor agents where hooks cannot be inserted. Source Uber Engineering