Google Cloud Security Blog · November 20, 2026

Google Cloud UAP preview: separate reusable policy rules from resource bindings

Why it matters

Google Cloud's Unified Access Policies preview combines allow and deny rules in reusable policy objects, with separate bindings determining where those rules take effect. This separates central policy authorship from local attachment and makes inherited access rules easier to manage across projects and folders. Creating a policy alone grants nothing: it must be bound, and a binding at a parent resource also affects descendants. The reference documentation adds important constraints to the launch story: existing allow and deny policies remain supported, custom roles cannot be used in UAP rules, and cross-organization bindings are not supported. It also recommends retaining existing policy types for established deployments, including their Policy Troubleshooter support. The launch post advertises 25 resource targets per policy, while the reference page lists five; that preview limit needs confirmation before designing a rollout. For agents, the useful pattern is independently enforced identity and permission scope, with tests covering inherited grants, explicit denies and exceptions.

My takeaway: Pilot one policy in a test project. Verify allowed and denied actions with the intended agent identity, test inheritance and exception principals, and confirm preview limits before expanding bindings.
Keep exploring

More curated notes connected through Agent Security and AI Compliance.

OpenAI News · framework

Pacing model development in an era of cyber-critical capabilities

OpenAI says preliminary evidence that Astra may meet its Critical cybersecurity threshold led it to pause frontier reinforcement-learning work for two weeks and keep its largest planned run on hold. New safeguards include stronger workload and network isolation, continuous boundary testing, token-level monitoring that escalates suspicious tool activity, and broader alignment checks for deception, reward hacking, and unauthorized access.

OWASP GenAI Security Project · guide

OWASP Top 10 for Agentic Applications for 2026

OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.

OpenAI News · framework

A blueprint for democratic governance of frontier AI

OpenAI proposes a three-part U.S. frontier-AI governance model: harmonize emerging state safety laws into a federal baseline, strengthen CAISI as an evaluation and standards institution, and coordinate a broader resilience program. Proposed controls include severe-risk evaluations, transparency reports, independent audits, safety-incident reporting, model-weight security, whistleblower protection, and periodic technical assessments.