{
  "schema_version": "1.6.0",
  "id": "ACVE-2026-0425",
  "aliases": [],
  "published": "2026-04-15T00:00:00Z",
  "firstReported": {
    "date": "2026-04-15",
    "url": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/",
    "publisher": "Aonan Guan"
  },
  "modified": "2026-09-25T00:00:00Z",
  "summary": "Text in a PR title, issue comment or hidden HTML comment hijacks an AI agent in GitHub Actions into leaking the job's secrets when it runs on pull requests or issues from untrusted authors.",
  "details": "## What\n\nAonan Guan with Zhengyu Liu and Gavin Zhong (2026-04-15) demonstrated it against Claude Code Security Review (`anthropics/claude-code-security-review`, via the PR title, leaking `ANTHROPIC_API_KEY` and `GITHUB_TOKEN`), Gemini CLI Action (`google-github-actions/run-gemini-cli`, via issue comments, leaking `GEMINI_API_KEY`) and the GitHub Copilot coding agent (via HTML comments in an issue body). Stolen values come back through PR or issue comments. Per the write-up, Anthropic's triage scored its case CVSS 9.3, then 9.4, then None; GitHub called it a known architectural limitation.\n\n## Detection\n\n`acve lock` cannot see CI workflows, so the matcher cannot fire on a developer machine: review `.github/workflows` by hand.\n\n## Fix\n\nDo not give agent jobs triggered by untrusted input any secret beyond what the task needs; require maintainer approval before they run.",
  "affected": [
    {
      "package": {
        "ecosystem": "GitHub Actions",
        "name": "anthropics/claude-code-security-review"
      },
      "ranges": [
        {
          "type": "ECOSYSTEM",
          "events": [
            {
              "introduced": "0"
            }
          ]
        }
      ]
    },
    {
      "package": {
        "ecosystem": "GitHub Actions",
        "name": "google-github-actions/run-gemini-cli"
      },
      "ranges": [
        {
          "type": "ECOSYSTEM",
          "events": [
            {
              "introduced": "0"
            }
          ]
        }
      ]
    }
  ],
  "references": [
    {
      "type": "REPORT",
      "url": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
    },
    {
      "type": "ARTICLE",
      "url": "https://www.securityweek.com/claude-code-gemini-cli-github-copilot-agents-vulnerable-to-prompt-injection-via-comments/"
    }
  ],
  "database_specific": {
    "severity": "HIGH",
    "acve": {
      "specVersion": 1,
      "status": "candidate",
      "kind": "code",
      "vulnClasses": [
        "prompt-injection-to-tool",
        "credential-exposure",
        "data-exfiltration"
      ],
      "cwe": [
        "CWE-1427"
      ],
      "exposure": {
        "harness": {
          "value": "A GitHub Actions job running anthropics/claude-code-security-review or google-github-actions/run-gemini-cli (the GitHub Copilot coding agent was also demonstrated)",
          "any": false,
          "status": "confirmed",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        },
        "model": {
          "value": "The default model of each action: Claude for the security review action, Gemini for the Gemini CLI action",
          "any": false,
          "status": "detected",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        },
        "goal": {
          "value": "Review the pull request for security issues, or triage and respond to the GitHub issue",
          "stated": true,
          "status": "confirmed",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        },
        "tools": {
          "value": "The agent's shell command execution; PR title, issue body and comments interpolated into the prompt; job secrets ANTHROPIC_API_KEY, GEMINI_API_KEY and GITHUB_TOKEN in the environment; PR and issue comments used to post results back",
          "status": "confirmed",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        },
        "approval": {
          "value": "The action runs headless with no human in the loop; the agent executes its tools automatically, and the demonstrated jobs set no allowed-tools restriction",
          "mode": "no-prompt-by-design",
          "status": "confirmed",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        },
        "inputControl": "content-author",
        "agentAction": "The agent follows instructions embedded in the PR title, issue body or comments and posts the job's secrets back as a PR or issue comment.",
        "harm": "credential-theft",
        "divergence": "goal-hijacked",
        "reach": {
          "value": "The CI job's secrets and tokens",
          "kinds": [
            "api-keys"
          ],
          "status": "confirmed",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        },
        "condition": {
          "value": "When the action runs on pull requests or issues from untrusted authors",
          "status": "confirmed",
          "source": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/"
        }
      },
      "occurrence": {
        "basis": "demonstrated",
        "reportedBy": "researcher"
      },
      "exploitation": {
        "status": "demonstrated",
        "checkedAt": "2026-09-21T00:00:00Z",
        "sources": [
          {
            "url": "https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/",
            "type": "research"
          }
        ],
        "kev": {
          "listed": false
        }
      },
      "taxonomy": {
        "owasp_asi": [
          "ASI01",
          "ASI03"
        ],
        "atlas": [
          "AML.T0093",
          "AML.T0051.001",
          "AML.T0086"
        ]
      },
      "cveBoundary": "user-configured",
      "noCveReason": "No CVE was assigned; the write-up reports bounties from Anthropic, Google and GitHub but no identifier.",
      "matcher": {
        "field": "harness.id",
        "op": "in",
        "value": [
          "claude-code-security-review",
          "run-gemini-cli"
        ]
      },
      "fix": {
        "summary": "Keep secrets out of agent jobs that untrusted PRs, issues or comments can trigger.",
        "actions": [
          {
            "type": "reconfigure",
            "target": ".github/workflows: triggers, permissions and secrets of the agent job",
            "why": "The agent treats PR titles, issue bodies and comments as instructions.",
            "owner": "operator"
          },
          {
            "type": "reconfigure",
            "target": "harness.github-action.allowedTools",
            "to": "no shell access unless the workflow grants it",
            "why": "Recommended: the demonstrated jobs set no allowed-tools restriction, and the agent posted the job's secrets back.",
            "owner": "harness-vendor"
          },
          {
            "type": "reconfigure",
            "target": "github.actions.untrustedTriggers",
            "to": "maintainer approval before a job with secrets runs on an outside author's PR or issue",
            "why": "Recommended: text from untrusted authors reached jobs holding the secrets.",
            "owner": "infra-provider"
          }
        ]
      },
      "reproducibility": {
        "status": "partial",
        "axesComplete": true,
        "componentsObtainable": null,
        "triggerPublished": false,
        "observableStated": true,
        "demonstrated": "researcher-demonstrated",
        "missing": [
          "model availability not checked (no registry reference)",
          "trigger not published"
        ]
      },
      "severityBasis": "harm-reach"
    }
  }
}
