AI Agent CI Security

Overview

• 2026-08-03: Coding-agent coverage in zenity platform (Claude/Codex/Cursor) + faster TS CI via typescript-7 (2026-08-03-zenity-series-c-official)

Security risks from integrating AI coding agents into CI/CD pipelines processing untrusted inputs — a distinct attack surface from model safety or prompt injection.

CVE-2026-12537 Case Study

gemini-cli CVSS 10.0 RCE (June 2026):

• Malicious .gemini/.env in fork PR workspaces
• Headless CI auto-trusts workspace configuration
• Pre-sandbox host-level code execution
• Affects google-github-actions/run-gemini-cli workflows

Pattern

AI agent supply-chain attacks via trusted configuration loading — parallel to agentjacking and langflow-cve-2026-33017-rce-exploited cluster.

Mitigation

• Upgrade gemini-cli ≥0.39.1, run-gemini-cli ≥0.1.22
• Never GEMINI_CLI_TRUST_WORKSPACE=true on untrusted PRs
• Minimal GitHub token permissions
• Explicit tool allowlists (v0.39.1 enforces under --yolo)

Related

• gemini-cli
• supply-chain-security
• ci-cd
• devsecops
• coding-agents
• github-actions

Sources

• 2026-06-29-google-gemini-cli-cve-2026-12537
• 2026-06-29-novee-gemini-cli-cve-research
• 2026-06-29-csa-gemini-cli-supply-chain-cvss10
• 2026-08-03-zenity-125m-series-c-ai-agent-security