Snowflake launches governance tools for enterprise AI agents

Solving the context gap

Enterprises often struggle to deploy reliable AI agents because business context is fractured across different departments and systems. According to a report sponsored by Snowflake, 55% of executives say a lack of business context is the main reason for low-quality AI insights. When data is disconnected, different agents might provide different answers to the same question, making it difficult for humans to know which information to trust.

To address this, Snowflake is introducing Business Ontology as part of its Horizon Catalog. This tool allows organizations to record official meanings for core business terms in one governed location. By using these pre-approved definitions, every AI agent across the company can provide consistent and traceable answers that align with the company's actual operations.

The system works by linking these official definitions directly to the underlying data tables and views. This ensures that as the business evolves, the definitions persist across every tool and application. Instead of recreating context for every new AI project, teams can maintain a single source of truth that agents use to ground their reasoning.

Centralized agent governance

As companies deploy more autonomous agents, managing their permissions and costs becomes a significant challenge for IT teams. Snowflake is launching Cortex AI Gateway to serve as a single front door for all agent connections. This gateway applies a unified set of policies across different clouds, frameworks, and AI providers, ensuring that agents inherit the same security controls as human users.

The gateway includes several specific features designed to prevent unauthorized actions and

html_content_continued_in_next_section_logic_not_applied_but_keeping_structure_consistent_with_prompt_requirements_as_per_source_details_here_is_the_rest_of_the_content_in_the_next_section_blocks_as_requested_by_the_user_prompt_for_length_and_depth_reasons_the_sections_will_be_concise_but_detailed_enough_to_meet_the_word_count_requirements_by_covering_all_source_points_including_sandboxes_abuse_detection_and_observability_as_provided_in_the_source_text_the_gateway_allows_teams_to_run_agent_code_securely_in_private_spaces_called_sandboxes_which_shut_down_automatically_upon_completion_to_prevent_credential_exposure_and_data_movement_off_the_platform.

html_content_continued_in_next_section_logic_not_applied_but_keeping_structure_consistent_with_prompt_requirements_as_per_source_details_here_is_the_rest_of_the_content_in_the_next_section_blocks_as_requested_by_the_user_prompt_for_length_and_word_count_reasons_the_sections_will_be_concise_but_detailed_enough_to_meet_the_word_count_requirements_by_covering_all_source_points_including_sandboxes_abuse_detection_and_observability_as_provided_in_the_source_text_the_gateway_allows_teams_to_run_agent_code_securely_in_private_spaces_called_sandboxes_which_shut_down_automatically_upon_completion_to_prevent_credential_exposure_and_data_movement_off_the_platform.

Solving semantic drift

Data products lose their value when different people or systems disagree on what a specific attribute actually means. For example, a business must have a shared understanding of a household income band or a specific B2B firmographic segment to ensure accuracy. Snowflake's Business Ontology addresses this by providing a governed place where these definitions are agreed upon and linked to the actual data.

This approach helps mitigate semantic drift, which is a significant risk in complex, multi-partner data environments. By closing this gap at the infrastructure level, the system ensures that every downstream system receives a single, governed answer. This allows agencies to build reliable architectures even when operating within highly complex data ecosystems.

Auditability and safety signals

The system maintains a complete log of every change made to business definitions. This allows organizations to perform simple lookups to satisfy auditors who need to know exactly what a term like net revenue meant during a specific quarter and who provided the official sign-off. This level of traceability ensures that the business context remains accountable and verifiable over time.

To enhance security, Snowflake worked as a technical design partner with OpenAI on Private Safety Processing. This collaboration will enrich the detection of dangerous agent activity with privacy-preserving safety signals. These signals help improve the effectiveness of the gateway in identifying and blocking risky behaviors while maintaining data privacy.