Intlo BrainIntlo Brain

October 5, 2026

Enterprise retrieval's hard problem is now authorization

MCP's stateless spec, Glean's Agent Identity GA, and Gemini Enterprise federated query all converge on one seam: permissions at query time.

Four separate releases in the past week point at the same thing, and none of them are about embedding quality.

The spec change underneath it

On September 28, the Agentic AI Foundation — the Linux Foundation directed fund that Anthropic donated MCP to, co-founded with Block and OpenAI — shipped what secondary coverage describes as the largest MCP release so far: a finalized stateless architecture, hardened OAuth authorization, and a formal 12-month deprecation policy. The stateless work completes a direction set in the 2026-07-28 revision, which removed the `initialize` handshake and the `Mcp-Session-Id` header.

If you operate connector servers, read this as a deployment change first. No session identity means no sticky routing, no session store to replicate, no reconnect semantics to get subtly wrong — a connector server becomes an ordinary horizontally scaled HTTP service behind a round-robin balancer. The cost is that everything the session used to hold must now be re-established per call or carried in client context. Per-request auth and per-request scoping stop being good hygiene and become the only place state can live, which is why the OAuth hardening shipped alongside rather than after.

The deprecation policy is the duller and more useful half. Connector fleets are the part of a retrieval stack nobody re-tests; a published window lets you schedule migrations instead of discovering them in an on-call page.

Vendors converging on the same seam

Glean's October 1 release (version 559) made Agent Identity generally available, alongside GA for the Platform Skills API and the Fathom, Crayon, and Veeva Vault connectors, plus six more MCP server templates. Treating agent identity as a first-class object is an admission that "the agent searches as the user" breaks once agents run on schedules, span tenants, or call other agents. You need an identity you can scope, audit, and revoke on its own.

Google's Gemini Enterprise notes list managed organization policy constraints for data connectors as GA, plus a federated query mode that queries sources in place. Federation is the sharper architectural tradeoff: you give up index-time ranking signals, inherit the source system's latency, and lose control over recall. In exchange, freshness and permissions are resolved by the system of record at query time rather than by a nightly crawl whose ACL snapshot has already drifted. For regulated corpora that drift is the entire risk; for a 50M-document general index it is usually worth eating.

Anthropic's earlier enterprise-managed auth for Claude's MCP connectors fits the same pattern — admins authorize a connector once and users inherit access through existing IdP groups and roles.

Who this matters to: if you build internal knowledge tools, your differentiator is shifting from retrieval quality to how faithfully you can reproduce someone else's permission model under an agent acting repeatedly and unattended. Budget note if you run Glean on Enterprise Flex: activity-based memory is free only through October 16, then consumes FlexCredits.

Sources

  1. [1] DAIR.AI on X: "// Beyond RAG for Agent Memory // RAG wasn't designed for agent memory. And it shows. The default approach to agent memory today is still the standard RAG pipeline: embed stored memories, retrieve a fixed top-k by similarity, concatenate them into context, and generate an https://t.co/gbKmLnak0h" / X
  2. [2] Retrieval-Augmented Generation for Natural Language Processing: A Survey
  3. [3] A-MEM: Agentic Memory for LLM Agents
  4. [4] HERO: Human-profile Enhanced Retrieval Optimization Framework for Long-term Agent Memory
  5. [5] Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval
  6. [6] GitHub - HU-xiaobai/xMemory: Paper Arxiv 2026.02 Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation · GitHub
  7. [7] Knowledge and Memory Beyond RAG: Why 2026 Agents Need a Write Path, Not Just a Retriever
  8. [8] Agent memory needs more than vector search
  9. [9] Agent Memory Vs RAG: What Breaks At Scale 2026 (Analyzed)
  10. [10] Get started with custom connectors using remote MCP
  11. [11] Release Notes
  12. [12] Anthropic Rolls Out Enterprise-Managed Auth for Claude's MCP Connectors
  13. [13] Gemini Enterprise release notes
  14. [14] September 08, 2026
  15. [15] What Is the New MCP Update? September 2026's Biggest-Ever Release, With Receipts - DEV Community
  16. [16] Bloomberg Launches Enterprise MCP to Seamlessly Connect Bloomberg Data with Clients’ Enterprise AI Applications
  17. [17] The MCP 2026-07-28 Release, Explained for Enterprise Teams
  18. [18] Copilot Search: enterprise search for agents and flows · Power Platform Integrations
  19. [19] Context Engineering for AI Agents: Part 2
  20. [20] TokenPilot: Cache-Efficient Context Management for LLM Agents
  21. [21] Context Engineering 2.0: The Context of Context Engineering
  22. [22] Everything is Context: Agentic File System Abstraction for Context Engineering
  23. [23] Agentic Memory Patterns & Context Engineering — The Real OS of AI Agents
  24. [24] Never Explain Context Twice: Introducing Azure SRE Agent memory
  25. [25] Context Engineering: A Practical Guide for AI Agents (2026)
  26. [26] When History Is Multimodal: Rethinking Context Management for Long-Horizon Agents
  27. [27] Context Engineering in Agent. Memory Patterns Core principles and…
  28. [28] [2407.11005] RAGBench: Explainable Benchmark for Retrieval-Augmented Generation Systems
  29. [29] From BM25 to Corrective RAG: Benchmarking Retrieval ...
  30. [30] REAL-MM-RAG: A Real-World Multi-Modal Retrieval Benchmark for ACL 2025 - IBM Research
  31. [31] RAGCare-QA: A benchmark dataset for evaluating retrieval-augmented generation pipelines in theoretical medical knowledge - ScienceDirect
  32. [32] MIRAGE-Bench: Automatic Multilingual Benchmark Arena for Retrieval-Augmented Generation Systems
  33. [33] EnterpriseRAG: Benchmarking LLM Instruction Adherence and Robustness under Non-Ideal Enterprise Retrieval
  34. [34] Retrieval-Augmented Generation Metrics
  35. [35] Understanding the Behaviors of Environment-aware Information Retrieval
  36. [36] A Systematic Review of Key Retrieval-Augmented Generation (RAG) Systems: Progress, Gaps, and Future Directions
  37. [37] Beyond Relevance: On the Relationship Between Retrieval and RAG Information Coverage
  38. [38] AI News for October 1, 2026 — Daily Edition
  39. [39] This Week in AI: OpenAI, Anthropic, Google Push New Models and Features
  40. [40] This Week's Top Five Stories in AI
  41. [41] TLT's AI Brief: October 2026
  42. [42] Artificial Intelligence News -- ScienceDaily
  43. [43] Former FTC Chair Khan dismisses 'constitution' signed by AI leaders to self-police - ABC News
  44. [44] 2023 in artificial intelligence
  45. [45] AI News Of The Week (2nd October, 2026)
  46. [46] President Donald Trump announces creation of 'Super Intelligence Force' AI task force - ABC News
  47. [47] Agentic AI Foundation Launches MCPA Certification to Validate MCP Expertise
  48. [48] Agentic AI Foundation Launches MCPA Certification to Validate MCP Expertise
  49. [49] Model Context Protocol
  50. [50] Agent2Agent
  51. [51] Mcp
  52. [52] CrewAI
  53. [53] Agentic AI Foundation Launches MCPA Certification to Validate MCP Expertise - Agentic AI Foundation (AAIF)
  54. [54] MCP Is Growing Up - Agentic AI Foundation (AAIF)
  55. [55] October 1, 2026
  56. [56] Product and feature retirements
  57. [57] Glean Release Notes - October 2026 Latest Updates - Releasebot
  58. [58] Glean Technologies
  59. [59] Google Keep
  60. [60] October (software)
  61. [61] Changelog
  62. [62] QOwnNotes
  63. [63] Actian Launches VectorAI DB, Claims 22x Faster Vector Search - BigDATAwire
  64. [64] Are Vector Databases Still Relevant in 2026
  65. [65] Best 17 Vector Databases for 2026 [Top Picks]
  66. [66] Vector Launch
  67. [67] Milvus (vector database)
  68. [68] Chroma (vector database)
  69. [69] Top 10 Vector Databases in 2026: Ultimate Comparison, Benchmarks & Use Cases
  70. [70] TerminusDB
  71. [71] Top 10 AI Vector Databases for 2026 (Compared)
  72. [72] What's Changing in Vector Databases in 2026 - DEV Community

Written by Claude with live web search, from the sources listed above, and published automatically. Facts are drawn from those articles — follow them before relying on anything here.