Microsoft enables the main Azure OpenAI access setting by default, while provider access, two cross-region processing controls and storage remain separate. Approval should record five settings across four decisions.

Montréal and Toronto are now supported for conversational analytics. The deployment decision is permanent per agent, and heterogeneous source locations trigger a global fallback unless teams choose a location explicitly.

Four enterprise products expose the same deployment trap: assistant configuration describes behavior, while permissions and data movement are enforced somewhere else.

AWS, Snowflake, Google Cloud and Oracle expose different failure boundaries around data, tools, identity and execution destinations. A deployable agent version must name all of them.

Cortex Agents defaults to continuing with accessible tools, but its authorization pre-check does not cover SQL, code execution or custom functions. Production clients need an explicit completeness policy.

Microsoft’s sharing model means an agent user can query a semantic model without Build permission or workspace membership. Review item-level grants, not just workspace roles.

Databricks now exposes a user’s group memberships to Genie One. Treat that context as a presentation and routing signal; keep authorization in Unity Catalog.

Google’s Data Agent Kit separates where Spark code runs from where notebook results can go. Security reviews need to test both boundaries.

Microsoft’s ALM format makes instructions, examples and schema scope reviewable. The catch: table selection is enforced, while column-level selection is not yet supported.

BIRD added a 74.51% hidden-test result with no development score or artifact link; Spider 2.0’s top rows expose a similar evidence boundary.

Cube’s production notes show why zero rows, truncated results and failed saves must be different machine-readable outcomes.

Cross-schema reuse narrows who can edit an agent team, while every consumer still needs the database and tool permissions required at runtime.

Thinking Machines has now identified the enterprise user behind ReViSQL-K2.6; the result is strong, reproducible and easy to misread.

watsonx.governance’s October update connects runtime controls, observability metrics and policy workflows. Teams still need to test whether the evidence proves the control worked.

Databricks, Google Cloud and Microsoft expose different shutdown boundaries for unattended analytics: code execution, orchestration identity and capacity consumption.

AWS’s internal Field Sales system treats data, tools, prompts and models as one releasable unit, then keeps testing after deployment.

PremiseSQL, CircleClar and EliteReplay measure different failure modes that execution accuracy collapses into one number.

Snowflake says semantic views and verified queries carry over unchanged. Teams should still regression-test routing, permissions and partial-tool access before switching production traffic.

Microsoft’s billing documentation separates model-token consumption from generated-query execution, while a worked example shows why teams should test both before rollout.

Three enterprise sources point to the same control-plane rule: identity cannot be bolted on only at query execution.

Promethium’s five-category framework is useful for procurement, but its own preferred category shows why buyers should separate architecture facts from vendor scoring.

Microsoft’s preview turns business definitions into searchable context. That makes retrieval quality a separate production test, not just a prompt-writing task.

The announcement removes a future cloud-placement objection, not a present one. Teams can connect Copilot now, but should separate client integration from compute and data residency.

The configuration details create two separate review points: who may execute generated notebook cells, and where scheduled workflows are actually sent.

The regional expansion removes a deployment constraint while leaving platform teams responsible for tightening the automatically provisioned gateway, logging and payload-capture settings.

Recent Snowflake, Microsoft, Immuta and Oracle guidance converges on a practical model: tool mode, database identity, session scope and lifecycle behavior must be governed separately.

Genie Code can parallelize legacy SQL conversion and repair syntax failures; teams still need result comparisons, review gates and versioned migration rules.

Version 1.21.7 makes automation changes confirmation-bound, limits subagent skills, fixes fail-open hooks and stops multi-statement SQL as a unit.

On Qwen2.5-Coder-7B, reward-specific token credit reached 73.15% average execution accuracy—1.60 points above GDPO—at a 10–20% training-step cost.

SQLcl MCP governs live database actions; Agent Memory governs durable context. Treating them as separate control planes makes identity, retention and failure testing much clearer.

A pending routing change improves sign-in locality, but enterprises with filtered public paths must discover, allow and test an additional account-specific hostname.

The public-preview design vends a warehouse role for each session, combining a human’s entitlements with an agent’s own access. That solves one problem while making agent policy design consequential.

The beta item-relations endpoints expose upstream and downstream dependencies as typed edges, giving coding agents a map for impact analysis while leaving change decisions outside the API.

The new RL recipe improves BIRD and Spider reach while resisting strategy collapse. Its strongest numbers measure repeated sampling, not leaderboard-style single-query accuracy.

The preview bundles database guidance by intent, but permissions—not the skill file—still decide what an agent can do.

Databricks’ new code-execution preview puts scheduled analysis on serverless sandboxes, but the safety decision is made from the task’s written scope on every run.

Across 16 models, explicit flaw warnings lifted aggregate detection to 93.1% but cut SQL generation on paired valid questions to 52.3%; the released code is inspectable, though the paper’s data and outputs are separate.

The connector is the headline, but project-scoped secrets and comment-safe query wrapping are the changes operators should regression-test first.

Kubernetes and S3-compatible connections widen the product’s reach, while credential masking and explicit read-only behavior show where that convenience needs hard boundaries.

Teams have until November 6 to move to Fabric IQ, gaining multi-agent catalog discovery and MCP endpoints while keeping cross-geo and identity constraints.

Guides give MCP-connected agents versioned business definitions close to the data. The harder task is deciding which metrics deserve central control and who owns each change.

The vendor’s new buyer guide is self-interested, but its proposed pilot checks are concrete: test identity passthrough, blocked writes, semantic grounding, concurrency and warehouse cost.

The 26.11 release couples Liveboard-grounded analysis with searchable conversations, tool-call traces and in-product accuracy tests. That makes agent governance operational, but also widens the data administrators can inspect.

The GA service exposes committed changes through low-lag views and preserves transaction boundaries, while making table-selection policy a new governance decision.

The new provider option reaches Google models through Oracle’s existing NL-to-SQL interface, while OAuth refresh tokens, session policy and regional endpoints create a new failure surface.

Microsoft now identifies agent prompts as background LlmPlugin operations, while the SQL they trigger appears on separate workload meters.

The official leaderboard verifies the score; Genloop’s site explains its context product, not enough of the evaluated system to reproduce the result.

Conversational analytics analyzes anonymized prompts by default, while the durable opt-out arrives as an organization-policy constraint on November 23.

The launch connects natural-language access, agent building and cross-system governance—but SAP’s announcement leaves runtime enforcement details to deployment design.

MongoDB says the payments company wants a reusable natural-language query pipeline, but its disclosed evidence is still an architecture decision—not a production result.