Snowflake and 1Password Spotlight AI Agent Security

According to SiliconANGLE's coverage of a theCUBE interview at Snowflake Summit, 1Password CTO Wang and Snowflake senior director Avanes discussed the rising operational and governance challenges posed by AI agents. Wang said, "The conversation has really changed from how do we secure systems to now how do we govern actors." Avanes said, "There is no world where you can just put an agent out or run your agentic tool directly against your data without those controls."

Per the 1Password blog dated April 30, 2026, 1Password expanded a partnership with Perplexity to integrate 1Password's secure access capabilities with Perplexity Computer. The 1Password blog quotes Perplexity CBO Dmitry Shevelenko: "To do that effectively in the enterprise, secure and seamless access has to be built into the experience from the start."

Per SiliconANGLE reporting from April 22, 2025, 1Password previously launched Agentic AI Security features and extended its XAM (Extended Access Management) platform to govern nonhuman and human access at machine scale, including an SDK for managing secrets in agentic workflows.

Industry-pattern observations: Enterprises are treating AI agents as a distinct class of actors that require runtime secrets handling, dynamic policies, and auditability rather than only traditional perimeter controls. That pattern increases demand for capabilities such as ephemeral credentials, least-privilege enforcement at runtime, and integration between identity systems and data platforms.

Industry-pattern observations: Snowflake's emphasis on tapping external metadata and semantic layers to make agentic workflows more accurate aligns with a broader move to couple data catalogs, schema metadata, and vector/semantic layers to improve agent grounding and reduce hallucination risk in production workflows.

Public reporting frames this set of announcements and discussions as part of a larger industry shift from question-answering to agentic, action-oriented workflows. The 1Password-Perplexity example used in the blog illustrates how agents can orchestrate multi-step tasks across portals while keeping credentials out of model inputs, a use case vendors are prioritizing to preserve confidentiality and compliance.

For data and security practitioners, the convergence of identity management and data governance raises operational complexity. Policymaking, logging, and telemetry need to cover machine actors, not just human users. That implies tighter integration between secret managers, IAM/policy engines, and data platforms that host or serve data to agents.

For practitioners: monitor how vendor SDKs and orchestration platforms handle ephemeral credentials, credential-less integrations, and fine-grained audit trails for agent actions. Watch for published integrations between secret stores (like 1Password's SDK) and orchestration/orchestrator platforms (like Perplexity Computer) and for Snowflake or other data platforms to publish prescriptive patterns or controls for agent access.

For observers: track whether enterprises adopt semantic layers or metadata-driven connectors to reduce agent errors, and whether audit and compliance tooling updates extend to nonhuman actors. Also watch for standards or best practices around agent identity, attribution, and runtime policy enforcement coming from major cloud and security vendors.

Reporting from SiliconANGLE and 1Password indicates vendor roadmaps and partnerships are converging on runtime secrets management, identity-data integration, and metadata-driven grounding as core components of AI agent security. These are practical priorities for teams building production agentic workflows.