authzed

Check out the February SpiceDB release. We introduced: -Postgres Foreign Data Wrapper -Query Planner -Keyword in Permissions Learn more yourself here, authzed.com/changelog/febr…

SpiceDB Community Day starts in exactly 8 minutes. Join us here: youtube.com/watch?v=UpwyIs…

AI is moving fast but most teams are still relying on authorization models that weren’t built for it. AI initiatives are stalling not because of technical complexity but because the permissions layer wasn't designed for it. We wrote more about why you need to think about modernizing authorization in the AI era here, authzed.com/blog/authoriza… If this is top of mind for you, join us today for SpiceDB Community Day where we will explore how to implement permissions for AI workflows (hear from Manuel de la Peña of @Docker talk about Permission-Aware RAG and Mark Fogle of @clawtocracy talk about how SpiceDB Stops Agentic Oversharing) and get a SpiceDB Update too. You can register or tune in here at 10 am PT, authzed.com/events/spicedb…

We are counting down the days to SpiceDB Community Day next Thursday at 10 am PT. We will discuss permissions in your AI workflow, SpiceDB updates and more. Don't miss hearing from: Manuel de la Pena from @Docker discussing Permission-Aware RAG: End-to-End Testing with the SpiceDB Testcontainer Mark Fogle from @clawtocracy discussing Selective Memory: How SpiceDB Stops Agentic Oversharing The AuthZed team will also be there to share the latest updates in SpiceDB (Query Planner, SpiceDB Foreign Data Wrapper for PostgreSQL and more) Register and join us, authzed.com/events/spicedb…

@AlexAI_Ontology Let us know if you have any questions. There are some self-guided workshops for learning SpiceDB here: github.com/authzed/worksh…

@authzed Oh, I’m setting up Codex and ReBAC. Codex recommended spiceDB, but I just did Neo4J. But Codex installed SpiceDB without me having time to stop it. I‘ll have to use it.

Prompts aren’t the core of AI work anymore. The real shift is structure. Build an ontology-centered system where AI can operate. Beyond simple vector chunking, connect data as nodes and edges with GraphRAG. Add ReBAC for access control. Then LangChain + MCP let agents act.

Excited to announce the @LangChain - SpiceDB integration library 🎉 This library brings SpiceDB's authorization model into LangChain & LangGraph workflows, so you can build Agentic and RAG workflows with fine-grained authorization that scales. Check it out: authzed.com/blog/langchain…

In case you missed the webinar, here's a recording youtu.be/PTWrOCyCfaI?si…

The AuthZed Cloud Datastore, Unlocked - As of today, we've unlocked the ability for customers of the AuthZed Cloud platform to scale operations themselves. AuthZed is stepping out of the way, not only to provide more visibility but to also give customers operational independence. Read more on the blog, authzed.com/blog/cloud-dat…, and try it out in AuthZed Cloud, authzed.com/products/authz…, today!

@TheNJDevOpsGuy @JamieMallers It's ReBAC with an option to add ABAC-style conditional expressions like time-based, IP-restricted access and so on. Some examples here: authzed.com/docs/spicedb/c…

@authzed @JamieMallers So SpiceDB can be for ReBAC and for policy enforcement based ABAC?

ReBAC and ABAC for Agentic authorization👇🏻 ✅ ReBAC Graph, relationship based, real-time authZ. Instead of defining roles (standard RBAC), you define a graph of relationships. Example: “Mike can view this document” might be true. Not because he has a ‘viewer’ role, but because he’s a member of the engineering team, which has access to the project folder, which contains the document. Platform to implement ReBAC: OpenFGA ✅ ABAC Handles the conditional request: Can “XYZ” be done based on the policy in place? It evaluates policies at runtime. These policies typically come from a policy enforcement tool (OPA, Kyverno, etc.). Because of the name (A stands for Attribute), you can think of attributes when creating ABAC on anything from geolocation to position to department to clearance and everything in between. Example: Mike has access to the production MCP Server tools, but he only has access to use them between the hours of 9:00-5:00. Both are making authorization decisions in real time, and both enforce policy. The real distinction is what data they use to make the decision. #kubernetes #ai #platformengineering

We're now live! 🔴 AI agent auth and what works today w/ Sohan Maheshwar of @authzed x.com/i/broadcasts/1…

@JamieMallers @TheNJDevOpsGuy SpiceDB uses ReBAC for "who can access what" and also ABAC-style caveats (contributed by Netflix) for relationships to be defined conditionally. In fact, OpenAI uses SpiceDB for their RAG usecase in ChatGPT Apps

@TheNJDevOpsGuy The combo is where it gets interesting. ReBAC for "who can access what" + ABAC for "under what conditions." Time-bounded MCP tool access is a pattern I expect to become standard — agents shouldn't have 24/7 prod access just because they were granted it once.

Model Context Protocol is does not secure by default. One of the big reasons is what @simonw calls the Lethal Trifecta. Watch this video to learn more about access control around MCP and how that caused some recent high-profile breaches youtube.com/watch?v=u0fUDB…

Access control will soon need to track: -Which documents an agent accessed -What an agent requested versus received -When permissions changed mid-task -How agent capabilities map to user permissions Agents will request access dynamically. Systems will make just-in-time decisions. Ambient context that constantly shifts as agents work and permissions update, doesn't fit the policy engine model. Policy Engines Don't Work for AI Authorization. authzed.com/blog/policy-en…

Cedar is not a good fit for AI Agent Authorization. That's because policy engines don't easily have access to what we like to term 'ambient context. See this video to learn more about why policy engines are NOT suited for AI Authorization youtube.com/watch?v=9HvE9T…

Building an AI agent is easy. Teaching it to keep a secret? 😅 At WebExpo, Sohan Maheshwar from @authzed scrutinises the leaks. 🤫 He architects permissions using the Google Zanzibar model to ensure your RAG system knows what to withhold. Live demo with Pinecone and SpiceDB.

Policy Engines are not a good fit for AI Agent authorization. Here's why that's the case, featuring an example that uses Cedar youtu.be/9HvE9Tm6Sss

Introducing the SpiceDB Foreign Data Wrapper (FDW) for PostgreSQL, a new experimental way to bring real-time authorization context from SpiceDB into Postgres queries, all without duplicating data, rewriting policies, or embedding authorization logic where it doesn't belong. Read more here and try it out yourself, authzed.com/blog/spicedb-p….

Introducing the #SpiceDB Query Planner (#iam #authz) - authzed.com/blog/introduci…

Everything is a relationship. Zanzibar, its origins, and how it led to SpiceDB were the focus of Sohan Maheshwar’s talk at Confitura 2025. Watch the recording here: youtube.com/watch?v=pC6Ggl…