Snowflake is introducing new AI security innovations, including Agent Identity for auditability, Horizon AI Guardrails for prompt injection protection, and Data Movement Policies to prevent unauthorized data exfiltration. These updates aim to provide enterprises with the controls needed to securely deploy autonomous AI agents at scale.
Snowflake is enhancing its platform with new AI security capabilities designed to secure agentic AI deployments in the enterprise. Key additions include Agent Identity (public preview), which provides a distinct signal to identify actions performed by an AI agent, enabling auditability and near real-time access restriction. Horizon AI Guardrails (generally available) offer prompt injection defense, integrated into the Horizon Catalog.
- Agent Identity
- Public preview
- Horizon AI Guardrails
- Generally available
- CoCo CLI Sandbox
- Private preview
- Data Movement Policies
- Private preview
- Data Exfiltration Detection
- Private preview
- Multi-Party Approval (MPA)
- Private preview
These innovations address the amplified security risks that come with autonomous AI agents making critical business decisions, such as malicious prompt injections and unauthorized data movement. The platform also introduces CoCo CLI Sandbox (private preview) for client-side isolation to mitigate data exfiltration and malicious code execution when AI systems run code.
Snowflake also provides Data Movement Policies (private preview) to prevent configured data movement from agents outside the Snowflake trust boundary. Additionally, Multi-Party Approval (MPA) (private preview) enforces a "four-eyes" rule for critical security operations, and new CoCo security skills simplify security administration through natural language.
