{
  "entry": {
    "kind": "vuln",
    "slug": "symjack-symlink-disguised-file-copy-tricks-ai-coding-agents-into-rce",
    "name": "symjack-symlink-disguised-file-copy-tricks-ai-coding-agents-into-rce",
    "vendor": "Symlink-Disguised File Copy Tricks AI Coding Agents into",
    "description": "Adversa AI disclosed SymJack, a symlink-disguised file-copy operation that tricks AI coding assistants into arbitrary code execution while the agent's approval prompt misrepresents what is actually being approved, the human approves one benign-looking action but a different, dangerous one runs. All six agents tested were vulnerable: Claude Code, Cursor, Antigravity, GitHub Copilot, and Grok Build. It is the latest instance of the recurring \"trust dialog\" class of agent weaknesses.",
    "website": "https://adversa.ai/blog/top-agentic-ai-security-resources-june-2026/",
    "repository": null,
    "documentationUrl": null,
    "license": null,
    "version": null,
    "trustLevel": "advisory",
    "capabilities": [],
    "packaging": null,
    "configFiles": null,
    "category": "agent",
    "updatedAt": "2026-05-26",
    "risk": {
      "score": 75,
      "band": "high",
      "signals": [],
      "dangerousCombos": []
    },
    "dangerousCombos": [],
    "vuln": {
      "id": "symjack-symlink-disguised-file-copy-tricks-ai-coding-agents-into-rce",
      "aliases": [
        "ti-031"
      ],
      "title": "SymJack: Symlink-Disguised File Copy Tricks AI Coding Agents into RCE",
      "summary": "Adversa AI disclosed SymJack, a symlink-disguised file-copy operation that tricks AI coding assistants into arbitrary code execution while the agent's approval prompt misrepresents what is actually being approved, the human approves one benign-looking action but a different, dangerous one runs. All six agents tested were vulnerable: Claude Code, Cursor, Antigravity, GitHub Copilot, and Grok Build. It is the latest instance of the recurring \"trust dialog\" class of agent weaknesses.",
      "severity": "high",
      "cvss": {
        "score": null,
        "vector": null,
        "version": null
      },
      "cwe": [],
      "attackType": [
        "Remote code execution"
      ],
      "componentType": "agent",
      "component": {
        "name": "Symlink-Disguised File Copy Tricks AI Coding Agents into",
        "vendor": null,
        "package": null,
        "ecosystem": null,
        "catalogSlug": null
      },
      "affected": [
        {
          "range": "6 AI coding agents tested (May 2026); all vulnerable",
          "fixed": null
        }
      ],
      "exploited": false,
      "exploitPublic": false,
      "publishedAt": "2026-05-26",
      "modifiedAt": "2026-05-26",
      "lastVerified": "2026-11-01",
      "references": [
        {
          "type": "article",
          "url": "https://adversa.ai/blog/top-agentic-ai-security-resources-june-2026/"
        }
      ],
      "mitigation": "Treat agent approval dialogs as untrusted when operating on cloned or untrusted repositories. Validate the real target of file operations before approving. Disable autonomous file-copy and symlink following where possible. Keep agents updated as vendors ship trust-dialog fixes.",
      "detectedByAnomity": true
    }
  }
}