The rapid advancement of autonomous AI agents capable of making critical business decisions introduces significant security risks alongside innovation. A recent Snowflake report highlights that 96% of businesses still face hurdles like data quality and skill gaps, underscoring the non-negotiable need for robust data security in AI deployments.
Security leaders are grappling with how to govern production-grade AI and defend enterprises against evolving threats. Snowflake is responding by integrating native, proactive, enterprise-grade security for data and AI workloads, aiming to give leaders confidence in deploying agentic applications at scale while maintaining data integrity and compliance.
Ankur Jain, Chief Cloud and Data Modernization Officer at Acxiom, noted the potential of Snowflake's new AI security capabilities to offer greater visibility and control over AI systems' access to sensitive data, facilitating responsible AI adoption.
Fortifying AI Workloads
Snowflake is bolstering its security portfolio across three core areas vital for AI success: agent security, data security, and platform-level security. This suite of features complements Snowflake Horizon Catalog, a central control plane for AI governance across all data.
Securing Agent Interactions
To manage agent identity and AI posture, Snowflake is introducing purpose-built controls. Agent Identity, currently in public preview, provides a distinct signal for AI agent actions, enabling auditability and near real-time access restrictions to sensitive data. This feature, alongside others, is part of Snowflake's effort to enable third-party agents to deliver advanced security solutions.
Prompt injection, a defining LLM security challenge, is addressed by Horizon AI Guardrails (generally available). This feature acts as a zero-day style defense layer integrated into the Horizon Catalog, offering granular, context-aware control over LLM interactions. This is crucial as AI Agents Demand New Security Rules.
For AI systems that run code, the CoCo CLI Sandbox (private preview) offers essential client-side isolation to mitigate data exfiltration and malicious code execution risks.
Simplifying Access Control
Frictionless and secure scaling of applications is being facilitated through Just-in-Time user provisioning and builder-initiated Request Access Workflows, both in private preview. The Access Troubleshooter skill within CoCo provides natural-language troubleshooting for access control errors.
Layered Defense for Data Estates
Snowflake is implementing a multi-layered approach to protect the entire AI data estate against unauthorized data movement, ransomware, and sophisticated exfiltration attempts. Data Movement Policies (private preview) aim to prevent configured data movement from Snowflake agents outside the trust boundary.
A new Data Exfiltration Detection package via Snowflake Trust Center includes anomaly detections for unusual data transfers and sensitive data fetches by agents. These are managed centrally through the Trust Center.
The daily brief
Get the free daily brief with the day's AI news, new AI startups and funding rounds every morning.
Join 64,000 other subscribers.
To combat ransomware and insider threats, Snowflake is introducing Multi-Party Approval (MPA) (private preview), enforcing a "four-eyes" rule for critical security operations. MPA can be combined with Tri-Secret Secure (TSS) for enhanced encryption control.
Snowflake Backups, introduced earlier this year, provide immutable point-in-time backups to protect against destructive actions, even by privileged users. Enhancements to account replication also support stringent recovery point objectives for mission-critical workloads.
Proactive Security Management
Snowflake Trust Center is evolving into a premier in-product AI security and compliance solution, offering centralized AI security posture management. This platform provides continuous visibility into security posture, displaying findings from Snowflake, third-party vendors, or custom detections.
AI-guided security posture management within Trust Center helps detect, prioritize, and remediate risks and misconfigurations with minimal setup. Proactive protections include malicious IP and leaked password detection.
Security management is being simplified through out-of-the-box CoCo skills. These skills empower administrators to perform critical security workflows using natural language, eliminating the need for complex UIs or manual scripts. Available skills include permissions analysis, security remediation, and network security.
Capitalizing on the agentic era requires a robust security foundation. Snowflake's integrated, AI-ready security capabilities aim to ensure that AI agents operate within centrally managed data policies and advanced defenses, maintaining the security and integrity of sensitive data.
