Release & Changelog Watcher — September 5, 2026

What Happened

• Amazon Bedrock Managed Knowledge Base: added a user-managed (3LO) authentication option for SharePoint, OneDrive and Confluence so teams can sign in with existing third‑party credentials for rapid prototyping; the prior service‑account (2LO) option remains for programmatic/production use [1].
• Amazon Bedrock Managed Knowledge Base: added a native ServiceNow data source connector that crawls knowledge articles, service catalog items and attachments, supports incremental syncs and can be scoped via sys ID inclusion lists [2].
• Amazon Bedrock Managed Knowledge Base: added automatic sync scheduling for native connectors (daily/weekly/monthly) so you can keep RAG indexes up to date without custom schedulers [3].
• Kubernetes v1.37 (released 2026-09-04): KubeletInUserNamespace (“rootless” node components) graduated to beta — node components can run as a non‑root host user via Linux user namespaces; feature gate enabled by default in beta but must be entered manually per node [4].
• AWS MCP Server (Agent Toolkit): added a serverless diagnostic capability for inspecting and troubleshooting running AWS Lambda functions and connected resources (API Gateway, EventBridge, S3, DynamoDB, SNS, SQS, Step Functions); diagnostics correlate 7‑day baselines and are available at no additional cost[5].

Why It Matters to Businesses

• Faster prototyping to production: Bedrock’s 3LO option reduces friction for business teams to connect enterprise content stores without waiting for IT to create service accounts — shortens time-to-proof-of-concept while preserving a 2LO path for secure production automation [1].
• Reduced engineering ops: native ServiceNow connector and scheduled syncs remove custom ingestion and scheduler maintenance, lowering ongoing engineering burden and accelerating delivery of IT/HR/customer-facing assistants that use live knowledge bases [2][3].
• Improved security posture options: Kubelet rootless beta gives teams another isolation layer for node components (useful for multi-tenant hosts, developer laptops and AI sandboxes) without changing pod behavior, enabling stricter host-user separation where needed [4].
• Faster automated debugging for serverless apps: AWS MCP Server’s Lambda diagnostics let AI coding agents retrieve consolidated configuration, change timelines and correlated error trends in one call — reducing agent complexity and token/API call overhead for runtime troubleshooting [5].
• Vendor lock‑in and governance considerations: convenience features that centralize crawling, indexing and diagnostics (Bedrock connectors, MCP) speed development but increase dependency on provider tooling and data flows that need governance and compliance controls.

Kimbodo Engineering Perspective

Practical trade-offs

Bedrock connectors (3LO vs 2LO): Use 3LO to enable product teams and business users to prototype assistants quickly; enforce 2LO/service-account flows for production systems that require programmatic credentials, auditable service identity, and least-privilege access. 3LO is higher convenience, 2LO is higher operational control [1].

ServiceNow connector and scheduled syncs: Native ingestion reduces pipeline engineering but requires careful scoping (sys ID lists), metadata mapping and sync cadence to avoid stale or noisy context in RAG. Use incremental syncs + scheduling to minimize cost and surface only authoritative KBs [2][3].

Kubelet rootless beta: Rootless node components lower blast radius on host account compromise but do not remove kernel-level escape risks. Treat this as an additional defense-in-depth control — pair with seccomp, AppArmor, cgroup limits and kernel hardening; test thoroughly on non-production clusters before adopting in critical environments [4].

AWS MCP Server diagnostics: Useful for rapid agent-driven troubleshooting, but requires IAM and access controls mapping. The centralized diagnostics (single-call aggregation) reduce token/API use; however verify that aggregated outputs don’t leak sensitive configuration or PII into agent contexts [5].

How We Would Implement It

Bedrock Managed Knowledge Base (SharePoint, OneDrive, Confluence, ServiceNow)

• Phase 1 — Prototype (days): enable 3LO connectors for business teams. Provide a self-service guide and template scopes; capture logs of which third‑party accounts connected. Use short-lived test assistants and synthetic data to validate relevance and retrieval behavior [1].
• Phase 2 — Hardening (weeks): switch production ingestion to 2LO service accounts that follow least-privilege IAM/third‑party roles. For ServiceNow, define sys ID inclusion lists and content filters; map metadata to your document schema and tag sensitive content for exclusion or redaction [2].
• Phase 3 — Sync cadence and monitoring: set automatic syncs by content volatility (daily for support KBs, weekly for policies, monthly for archives), enable incremental syncs, and monitor sync error rates and document churn; log provenance and snapshot vectors for auditability [3].
• Operational controls: integrate Bedrock connectors with your identity provider, require MFA for 3LO sign-in, store audit logs in a secure SIEM, and enforce retention/PII rules in the ingestion pipeline.

Kubernetes v1.37 rootless Kubelet

• Test plan: create a dedicated test cluster using kind or minikube with a rootless runtime (rootless Docker/nerdctl/Podman). Verify kernel and sysctl prerequisites, and confirm node YAML shows runningInUserNamespace for nodes in the namespace [4].
• Deployment: enable feature gate on non-production clusters, run kubelet and kube-proxy as a mapped non-root host UID inside a pre-created user namespace, and validate mounts, cgroups and CNI behavior. Document required systemd and kernel settings for reproducible bootstrap.
• Operational controls: retain traditional host security (seccomp, AppArmor, kernel updates). Monitor for permission-related errors tolerated by the beta and contribute feedback upstream where practical.

AWS MCP Server — Lambda diagnostics

• Enablement: install/configure the Agent Toolkit or enable the MCP Server and run “aws configure agent-toolkit” per AWS guidance; ensure the MCP Server has the minimum IAM permissions needed to read deployed configs and logs [5].
• Integration: wire diagnostics into your AI coding agents (restrict outputs to non-sensitive summaries), consolidate diagnostics into your incident workflows, and log diagnostic requests and returned artifacts to your audit store.
• Governance: restrict MCP Server access to approved operator roles, configure region awareness (MCP runs in us‑east and eu‑fra), and enforce redaction of secrets before using diagnostics in chat contexts.

Risks, Costs and Security

• Data exposure and PII: Connectors and diagnostics aggregate enterprise content. Mitigation: implement content classification pre-ingest, apply redaction/PII policies, and prevent LLM context windows from storing or surfacing raw secrets. Audit all connector sign-ins and sync activity [1][2][5].
• Access management trade-offs: 3LO is convenient for prototyping but increases the number of human credentials with access paths. Require MFA, limit consent scopes, and transition production workloads to 2LO service accounts for auditability [1].
• Operational cost vs. convenience: Scheduled syncs reduce engineering overhead but increase storage and compute for incremental indexing. Choose cadence by content volatility; monitor sync costs and throttling [3].
• Kubernetes rootless caveats: Rootless Kubelet reduces host UID blast radius but does not mitigate kernel vulnerabilities, idmapped tmpfs and cgroup edge cases remain; beta may change behavior before GA — avoid rolling into critical clusters without validation [4].
• Centralization and vendor lock-in: Relying on provider-managed crawlers and diagnostics accelerates delivery but can increase dependency on provider APIs and data formats. Mitigation: maintain exportable raw data backups and an independent ingestion fallback for portability.
• Audit and compliance: Ensure connector sync logs, MCP diagnostic calls and user consent records are forwarded to SIEM/archival systems; use role‑based restrictions to limit who can query aggregated diagnostics and RAG contexts [1][2][5].

Sources

1. [1] Amazon Bedrock Managed Knowledge Base introduces user-managed setup for SharePoint, OneDrive, and Confluence data sources
2. [2] Amazon Bedrock Managed Knowledge Base now supports ServiceNow as a native data source connector
3. [3] Amazon Bedrock Managed Knowledge Base now supports automatic sync scheduling for data source connectors
4. [4] Kubernetes v1.37: KubeletInUserNamespace (aka Rootless mode) Graduates to Beta
5. [5] AWS MCP Server adds a serverless capability for AWS Lambda functions