Hey everyone! 👋 hope your week is off to a good start! We just published a page on AI gateway authorization, walking through how to layer fine-grained access control on top of your AI gateway. Feel free to check it out if it’s relevant for you ☺️ AI gateways are good at telling you who’s calling but not what that caller is actually allowed to do. Once an agent starts using tools, calling an LLM, or delegating to another agent, the human behind the request tends to disappear from the audit trail, sub-agents end up inheriting more privilege than they should, and revoking access usually means rotating credentials or redeploying. The page covers what changes when every model call, tool invocation, and agent-to-agent handoff runs through the same policy layer your apps already use. You end up with one audit trail across humans, services, and agents, sub-agents that stay scoped to what the parent had, and policy changes that take effect at request time rather than at deploy time. Plus evidence trails that line up with SOC 2, ISO 27001, HIPAA, and DORA.