AIToday
Large Language ModelsAI Coding AssistantsAI Safety & AlignmentZenn AI/MLPublished: Oct 8, 2026, 10:00 JST

PERSON + Snowflake OAuth(IS_AGENTIC = TRUE): best way to connect agents

PERSON + Snowflake OAuth(IS_AGENTIC = TRUE): best way to connect agents

3 Key Points

  1. What happened

    Writing on Zenn, NOWCAST data engineer Kebin compared ways to connect Snowflake's Agent Identity to coding agents and concluded PERSON + Snowflake OAuth(IS_AGENTIC = TRUE) is the only fit.

  2. Why it matters

    The author warns that passing a human user's key pair or PAT to an agent leaves Snowflake unable to tell human from agent, audit who the agent acted for, or limit an agent takeover.

  3. What to watch

    On OAuth integration creation, OAUTH_REFRESH_TOKEN_VALIDITY is set to 28800 seconds and OAUTH_ACCESS_TOKEN_VALIDITY to 3600 seconds.

WHO IT HITSData platform and security engineers running coding agents against Snowflake will need to stop sharing human credentials and move agents onto a dedicated OAuth integration to keep audit trails and least-privilege controls intact.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The article opens with the practical problem at NOWCAST: engineers run coding agents inside DevContainers rather than directly on local PCs and limit mounted credentials, but if the Snowflake credential handed to the agent belongs to a human, Snowflake still sees a human session. That leads to four concrete gaps: no agent identification, excessive inherited privileges, long-lived key pairs or PATs, and an inability to disable only the agent without stopping human work.

Agent Identity is described not as a single feature but as a group of capabilities built around one fact, whether a session is agent-initiated. Once Snowflake marks a session agent-active, that fact feeds into audit views, data-protection policies, and control limits. The article distinguishes delegated sessions (a user's session marked agent-active) from autonomous ones (a dedicated SERVICE_AGENT user), noting the outcome depends on which user the agent connects as, not on the agent type. A setup example shows a DevContainer-based Snowflake CLI obtaining a token through an agentic OAuth integration, with the session then recorded as EXTERNAL_AGENT and restricted from switching roles.

FAQ
How does the recommended setup prevent agents from using admin roles?
The OAuth integration sets ALLOWED_ROLES_LIST to a specific role and BLOCKED_ROLES_LIST to roles including ACCOUNTADMIN, SECURITYADMIN, SYSADMIN, and USERADMIN. It also sets OAUTH_USE_SECONDARY_ROLES = NONE so secondary roles cannot be used.
What makes Snowflake OAuth with IS_AGENTIC = TRUE the author's preferred choice?
The author says it is the only option that meets all four requirements—agent identification, limited roles, short-term credentials, and separation from humans—and lets auditors trace which human the agent acted for.
How can policies be tested without starting a real agent?
The article shows that POLICY_CONTEXT can simulate an agent session by running EXECUTE USING POLICY_CONTEXT with SNOWFLAKE$CURRENT_ACTIVATED_AGENT_TYPES => ('EXTERNAL_AGENT'). This returns IS_AGENT_ACTIVATED = TRUE and AGENT_TYPE = EXTERNAL_AGENT.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleAI cluster networking reading list targets GPU engineers