Pull to refresh
Logo
What Changed
Following Why Sign Up
Snowflake moves AI agent governance into its data catalog

Snowflake moves AI agent governance into its data catalog

New Capabilities

Business Ontology and Cortex AI Gateway aim to make agents answer to approved definitions and existing access rules

Yesterday: Agent controls detailed for customers

Overview

Updated 2 hours ago

Snowflake built its business on holding other companies' data. At its Expedition 2026 conference, it started selling the controls for the AI agents that read that data.

Two products carry the pitch. Business Ontology is a shared dictionary of approved business terms, tied to the tables and metrics that calculate them. Cortex AI Gateway is a single checkpoint for every model call and tool call an agent makes, applying the access rules that already govern human employees.

The bet is that governance belongs where the data sits, not in a separate security product added later. If buyers agree, Snowflake becomes the place companies decide what an agent may know and do. If they don't, the gateway is one more console beside Microsoft's, Google's and Databricks'.

Why it matters

Agents can now move money and edit records. Snowflake wants the permission check to happen in the warehouse, not the agent.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

2
New governance products
Business Ontology and Cortex AI Gateway, both housed in Horizon Catalog.
6
Control areas in Cortex AI Gateway
Sandboxes, abuse detection, observability, inference guardrails, cost caps and tool-call checks.
2
Agents grounded in the ontology
CoCo for data teams and CoWork for knowledge workers.

Voices

Curated perspectives — historical figures and your fellow readers.

Sign up to generate historical perspectives on this story.

Log in Sign Up

People Involved

Sridhar Ramaswamy
Sridhar Ramaswamy
Chief executive, Snowflake
Leading the company's AI and governance push

Organizations Involved

Snowflake Inc.
Snowflake Inc.
Cloud data platform
Vendor behind Business Ontology and Cortex AI Gateway

Snowflake sells cloud data storage and computing, and now sells governance for AI agents that run on that data.
OpenAI
OpenAI
AI model developer
Technical design partner on agent safety detection

OpenAI worked with Snowflake on privacy-preserving safety signals for detecting dangerous agent behavior.
Databricks
Databricks
Cloud data and AI platform
Competitor selling a rival governance layer

Databricks sells a competing data platform whose Unity Catalog plays a similar governance role.

Timeline

2012 November 2026

5 events Latest: Yesterday
Tap a bar to jump to that date

  1. Agent controls detailed for customers

    Latest Product launch

    Snowflake describes sandboxes, abuse detection, trace logs, cost caps and tool-call checks for agents.

  2. Expedition 2026 opens with governance pitch

    Conference

    Snowflake announces Business Ontology and Cortex AI Gateway, both inside Horizon Catalog.

  3. Ramaswamy becomes chief executive

    Leadership

    The former head of Snowflake's AI division replaces Frank Slootman.

  4. Snowflake goes public

    Company

    The initial public offering makes Snowflake one of the most valuable software companies to list.

  5. Snowflake founded

    Company

    The company starts as a cloud data warehouse built for separating storage from computing.

Scenarios

1

Business Ontology reaches general availability

Likely Resolves by Q2 2027

Discussed by: Snowflake's own release plan, which lists the feature as entering public preview soon

Snowflake said Business Ontology would enter public preview soon, with general availability to follow. The trigger is customer demand for one definition layer that agents can query instead of rebuilding terms for each tool. If preview customers struggle to map legacy glossaries, the date slips.

2

Named customers credit the gateway for clearing security review

Possible Resolves by Q2 2027

Discussed by: Snowflake's launch materials, which frame the gateway as a precondition for production agents

The commercial test is whether security teams accept the gateway as the control point instead of building their own. Watch for named companies in Snowflake case studies or earnings calls saying the gateway let them approve an agent for production.

3

A rival bundles equivalent controls and the gateway stops being a reason to buy

Possible Resolves by Q2 2027

Discussed by: TechTarget's reporting, which notes the features still need proof at scale

Microsoft, Google and Databricks all sell some form of agent governance. If one of them includes data-layer agent controls at no extra cost, Snowflake's gateway becomes a feature rather than a product. The trigger is a rival announcement that prices the capability at zero.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

September 2016

Google buys Apigee (2016)

As companies exposed internal systems through APIs, a gateway that checked identity, rate limits and logging became the standard place to enforce policy. Google bought Apigee, one of the largest independent gateway vendors, that September.

Then

Cloud providers bundled API management into their platforms.

Now

Gateways became default infrastructure, and the independent category consolidated into cloud suites.

Why this matters now

Agent gateways are following the same path. A new kind of traffic needs a checkpoint, and platform vendors are racing to own it.

April 2011

Model risk management guidance (2011)

The Federal Reserve and the Office of the Comptroller of the Currency issued guidance requiring banks to inventory and validate the models they relied on, including models bought from vendors.

Then

Banks built model risk teams and formal inventories.

Now

Validation before deployment became a standard part of bank supervision.

Why this matters now

Agent inventories and trace logs are the same move: prove what the system does before it does it.

May 2018

EU data protection rules take effect (2018)

The General Data Protection Regulation required organizations to document what personal data they held and why before processing it.

Then

Companies built data maps and hired privacy officers.

Now

Documentation became a procurement requirement, not a compliance afterthought.

Why this matters now

Business Ontology is a documentation layer of the same kind, applied to business definitions instead of personal data.

Sources

(5)
+2
Snowflake TechTarget StreetInsider Snowflake Snowflake