Snowflake is adding a new enterprise AI tool aimed at making model customization easier for businesses that want more control over their own data and workflows.

At its annual Snowflake Summit in San Francisco, the data cloud company introduced Cortex Training, a service designed to let organizations fine-tune open-weight models using their own proprietary information. The move is part of a broader set of AI updates Snowflake announced this week, all centered on helping enterprises build and use AI systems without sacrificing security, governance, or access to business context.

The company said the new capability is meant to help customers adapt models to specific enterprise needs while keeping sensitive information inside Snowflake’s governed environment. That matters for companies that want AI tools tailored to internal terminology, processes, and records, but do not want to move data across multiple systems or expose it to unnecessary risk.

Snowflake also expanded several other parts of its AI platform. Snowflake CoWork, previously known as Snowflake Intelligence, is being positioned as a personal AI agent for employees. The platform combines data access, business context, and tools in one place. New features include Cortex Sense, which combines data with business definitions and operational knowledge, User Memory, which can help people schedule tasks and take actions through integrations such as Gmail and Slack, and a Skill Catalog for discovering and reusing agent capabilities.

The company also updated Horizon Catalog, its hub for AI governance, context, and security. Snowflake said the catalog is intended to make sure people, tools, and AI agents are working from the same business definitions. It also includes features such as Adaptive Compute, verified agent identities, continuous security posture management, and protections against jailbreak attempts and zero-day vulnerabilities.

On the developer side, Snowflake renamed its coding assistant Snowflake CoCo, formerly Cortex Code, and said it can now complete tasks autonomously without requiring a user to remain on screen. The company also introduced Datastream, a feature meant to give AI applications ongoing access to current data without the need for brokers, connectors, or separate streaming infrastructure.

Snowflake executives framed the announcements around a familiar challenge for enterprise AI: balancing usefulness with trust. Christian Kleinerman, the company’s executive vice president of product, said Snowflake’s goal is to help organizations make workers more productive while giving them confidence in security, compliance, and governance.

The company also highlighted continued work with Anthropic, whose Claude models are available on Snowflake Cortex AI. Anthropic president Daniela Amodei said the partnership reflects the idea that trust can speed adoption in business settings. She added that enterprise customers are looking for dependable systems rather than ones that produce more errors or hallucinations.

Taken together, the updates show Snowflake leaning further into the role of AI infrastructure provider for large organizations. Rather than competing as a consumer-facing AI chatbot, the company is trying to give businesses a controlled environment for connecting their data, policies, and applications to frontier models. Cortex Training extends that strategy by giving enterprises another way to adapt open-weight AI models to their own needs while keeping oversight in place.