Wiz Research's "Red Agent"—an autonomous, AI-powered security tool—identified a critical GitHub Actions workflow vulnerability in Snowflake's public repository snowflakedb/snowflake-connector-net. The flaw allowed unauthenticated users to execute arbitrary commands within a GitHub Actions runner by opening an issue with a crafted title. The vulnerable pattern was introduced on June 18, 2026, when PR #1218 was merged, co-authored by "Copilot Autofix powered by AI." The merged PR replaced a safe input sanitization method with direct string interpolation, creating a script injection vector. Despite GitHub's AI-assisted security review flagging the change as safe, the vulnerability remained undetected.

Upon discovery, Wiz responsibly disclosed the issue to Snowflake on June 23, 2026, via HackerOne report #3819931. Snowflake immediately patched the workflow (commit 1dc7766, PR #1402), rotated the compromised Jira token, and verified through audit logs that Wiz was the sole actor during the five-day exposure window. All data accessed during proof-of-concept testing was securely deleted.

The incident highlights risks associated with AI-assisted code generation, where probabilistic predictions can inadvertently reintroduce insecure coding patterns. Wiz's Red Agent demonstrated adaptive exploitation capabilities, autonomously adjusting its payload after an initial syntax error to successfully exfiltrate credentials. The exposed token authenticated as [email protected], granting read access to Snowflake's engineering and security compliance Jira projects.