Snowflake is taking aim at one of the biggest challenges facing enterprise artificial intelligence: AI agents that can access company data but struggle to understand what that information actually means.
At Snowflake Expedition 2026, held on November 3, the company announced new capabilities for Snowflake Horizon Catalog, including Business Ontology and expanded Cortex AI Gateway controls. The technologies are designed to help organizations deploy AI agents with more consistent business knowledge, stronger security and better oversight of automated operations.
The problem goes beyond AI accuracy. As businesses connect autonomous agents to internal databases, applications and external AI models, questions about accountability, access permissions and unexpected costs are becoming harder to ignore.
Snowflake believes the answer lies in giving AI agents the same approved business definitions and governance rules that organizations already use to manage their data.
Snowflake Business Ontology Gives AI Agents a Shared Understanding of Company Data
Different departments often use the same business terms to describe different things. Revenue might refer to gross sales in one report and net revenue in another. Customer activity can carry different meanings across marketing, finance and operations. For an AI agent trying to answer questions across these systems, the confusion can produce inconsistent results.
Snowflake’s new Business Ontology aims to address that issue by creating a governed source of business definitions. Instead of allowing AI systems to interpret terminology independently, organizations can establish approved meanings for important concepts and connect those definitions directly to their underlying data.
The technology operates within Horizon Context, part of Snowflake Horizon Catalog. It allows businesses to identify who owns a definition, track its approval status and maintain relationships between business concepts and the data used to calculate them.
For example, a finance department could approve a specific definition of net revenue and link it to the relevant datasets. AI agents working with that information would then use the approved definition rather than generating their own interpretation.
Business Ontology also works alongside Snowflake Semantic Views and Cortex Sense. Semantic Views provide the logic used to calculate business metrics, while Cortex Sense helps identify additional context from metadata, queries and usage patterns. Together, these capabilities are intended to make AI-generated answers more consistent with how a company actually operates.
Snowflake expects Business Ontology to become available in public preview soon.
Cortex AI Gateway Brings Enterprise AI Agents Under Centralized Governance
As companies introduce more AI agents into their operations, managing their connections to different applications and models becomes increasingly complicated. An agent might retrieve financial information, interact with a customer database or execute a task through an external service. Each interaction introduces another point where security policies and permissions need to be enforced.
Snowflake is expanding Cortex AI Gateway to provide a centralized control layer for these activities.
The gateway allows organizations to govern how AI agents access models, communicate with enterprise tools and perform actions across different cloud environments. Rather than establishing separate policies for every agent, companies can apply existing authentication and authorization rules to AI-driven workflows.
The system also supports governance of Model Context Protocol (MCP) connections, which allow AI agents to interact with external tools and services. By checking tool requests against established permissions, Cortex AI Gateway is designed to prevent unauthorized actions before they reach sensitive systems.
For businesses deploying agents across multiple departments, centralized oversight could reduce the complexity of managing AI access. It also provides a way to examine which systems agents interact with and whether those interactions comply with company policies.
Snowflake Adds Security Monitoring to Detect Risky AI Agent Behavior
AI agents are becoming more autonomous, but greater independence creates new security concerns. An agent that receives an inappropriate instruction or attempts an unauthorized action could expose information or disrupt business processes.
Snowflake is introducing Agent Abuse Detection to help organizations identify potentially dangerous activity.
Unlike systems that examine individual prompts in isolation, the capability analyzes sequences of agent actions. This approach is intended to help security teams identify suspicious patterns, including attempts to bypass permissions, access restricted information or misuse enterprise tools.
The company is also developing Agent Trace Intelligence, which allows organizations to examine how agents perform tasks and investigate problems such as repeated tool failures, inefficient processes and unexpected behavior.
Snowflake said it has worked with OpenAI as a technical design partner on Private Safety Processing, a developing approach intended to provide additional privacy-preserving safety signals for identifying dangerous agent activity.
Another feature, Cortex Sandboxes, provides isolated environments where AI agents can execute custom code while following the permissions of the person initiating the task. The aim is to support automated workflows without unnecessarily exposing credentials or transferring sensitive data outside Snowflake.
Several of these security capabilities remain in private preview or development, so their availability will vary.
AI Cost Management Becomes a Priority for Enterprise Adoption
AI spending can become difficult to predict when organizations deploy multiple agents across different departments. An automated process that repeatedly calls an expensive model or gets stuck performing the same task could generate substantial costs before anyone notices.
Snowflake is addressing this problem through additional cost governance capabilities within Cortex AI Gateway.
Organizations can attribute AI spending to individual users, departments, applications and agents. This provides administrators with a clearer understanding of where resources are being consumed and which workloads account for the largest expenses.
The platform also supports spending limits intended to stop automated processes from exceeding approved budgets.
Snowflake’s Dynamic Model Routing capability takes a different approach to cost optimization. Rather than sending every request to the same AI model, it is designed to select an approved model based on factors such as cost, accuracy and response time.
A routine classification task, for instance, might not require the same computational resources as a complex analytical request. Directing workloads to suitable models could help companies reduce unnecessary spending while maintaining acceptable performance.
Dynamic Model Routing remains in private preview.
Research Highlights Why Business Context Matters for AI Accuracy
Snowflake’s announcement arrives as businesses continue struggling with the reliability of AI-generated insights.
According to research conducted by MIT Technology Review Insights and sponsored by Snowflake, 55% of surveyed executives identified insufficient business context as the primary reason AI agents produce poor-quality insights. Another 45% pointed to fragmented or inaccessible data.
The survey involved 400 senior data and technology executives from organizations generating at least $500 million in annual revenue.
Those findings suggest that improving AI models alone may not solve every problem businesses encounter when deploying autonomous systems.
An AI agent might retrieve accurate figures from a database but still deliver a misleading answer if it misunderstands the company’s reporting rules. The same problem can occur when different departments use conflicting definitions or maintain separate versions of important business information.
Snowflake’s Business Ontology is designed to reduce those inconsistencies by giving AI agents access to definitions that organizations have already reviewed and approved.
Glean, Kargo and dentsu Explore Snowflake’s Enterprise AI Capabilities
Several organizations are already working with Snowflake’s governance and business context technologies, offering early examples of how the capabilities could support enterprise AI deployments.
Glean is using Cortex AI Gateway to strengthen visibility and control across AI interactions as agents become more sophisticated and operate across additional enterprise systems.
Kargo has highlighted the importance of understanding which teams and agents are responsible for AI spending. By using Cortex AI Gateway, the company can examine consumption patterns, establish spending limits and direct requests toward models suited to particular workloads.
Meanwhile, dentsu and its data and technology agency Merkle are early adopters of Business Ontology.
For businesses managing complex customer information, advertising analytics and data partnerships, inconsistent definitions can create problems across reporting systems. Business Ontology gives teams a way to connect approved terminology to the information their applications and AI agents use.
Snowflake is also collaborating with Uniphore to extend governance capabilities into broader enterprise AI workflows.
These deployments reflect growing interest in managing AI agents as part of established business infrastructure rather than treating them as isolated experiments.
Snowflake Wants Governance to Become the Foundation of Agentic AI
The enterprise AI market is moving beyond basic chatbot applications. Businesses increasingly want systems capable of analyzing information, coordinating workflows and completing tasks with limited human intervention.
That shift raises expectations for reliability.
Companies need to understand not only what an AI agent produces, but also which information it used, which actions it performed and whether those actions followed established rules.
Snowflake is positioning Horizon Catalog as a foundation for meeting those requirements. Business Ontology focuses on how AI agents interpret business information, while Cortex AI Gateway governs their interactions with models, applications and enterprise systems.
The combination could help organizations address two persistent barriers to AI adoption: inconsistent answers and limited operational control.
There are still practical questions surrounding implementation. Businesses will need to maintain accurate definitions, assign responsibility for approving changes and configure security policies carefully. Governance tools can reduce risks, but they cannot guarantee that every AI-generated answer or autonomous action will be correct.
Even so, Snowflake’s latest announcement points toward a larger change in enterprise AI.
As autonomous agents become part of everyday business operations, the ability to control their behavior and verify their decisions may prove just as important as the intelligence of the models powering them.
For Snowflake, that creates an opportunity to make governed enterprise data central to the next phase of AI adoption.
Sources
- Snowflake — Official Press Release, November 3, 2026: Snowflake Helps Enterprises Deploy AI Agents with Trusted Business Context and Built-In Controls
- Snowflake — Product and Technology Blog: Business Ontology Governs Business Concepts and Metrics for AI and Analytics
- Snowflake — Product Information: Snowflake Cortex AI Gateway
- TechTarget — November 3, 2026: Latest Snowflake Blast Targets Trust in AI and Efficient Development
