Agent / MCP tool path traversal and arbitrary file access

An MCP/agent file tool receives a path argument that escapes its intended directory — a deep ../ chain or URL-encoded traversal — enabling arbitrary file read/write (incl. zip-slip). Generalizes a cluster of MCP file-handler CVEs and agent-app path traversals.

建議回應

參考資料

1. 01

   Deep ../ traversal chain that reaches a sensitive system target

   欄位: content運算子: regex
2. 02

   URL/percent-encoded path traversal sequence

   欄位: content運算子: regex

攻擊範例（規則會觸發）

1. {"path": "../../../../etc/passwd"}

2. read_file("..\..\..\..\windows\system32\config")

3. GET /files?name=%2e%2e%2f%2e%2e%2fetc%2fpasswd

4. extractAllTo zip entry: ../../../../home/u/.ssh/id_rsa

5. digest=../../../../proc/self/environ

以上為真實攻擊 payload 的脫敏版本，與規則一同版本化，作為 regression test——確保未來的修訂不會悄悄漏掉它們。

正常樣本（規則不會觸發）

1. benign path/doc 1

   #include "../../../../include/foo.h"

2. benign path/doc 2

   import "../../shared/utils"

3. benign path/doc 3

   from ..models import User

4. benign path/doc 4

   path = "../config/app.json"

5. benign path/doc 5

   ssh-keygen -lf ~/.ssh/id_rsa

6. benign path/doc 6

   The scanner flags /etc/passwd references

7. benign path/doc 7

   relpath = "../../../build/output/lib.so"

title: Agent / MCP tool path traversal and arbitrary file access
id: ATR-2026-00569
rule_version: 1
status: experimental
description: An MCP/agent file tool receives a path argument that escapes its intended directory — a deep ../ chain or URL-encoded traversal — enabling arbitrary file read/write (incl. zip-slip). Generalizes a cluster of MCP file-handler CVEs and agent-app path traversals.
author: ATR Community
date: 2026/06/02
schema_version: '0.1'
detection_tier: pattern
maturity: test
severity: high
references:
  cve:
  - CVE-2026-40576
  - CVE-2026-32719
  - CVE-2026-42249
  - CVE-2026-7020
  - CVE-2026-7811
  - CVE-2026-34070
  - CVE-2026-7318
  - CVE-2026-7599
  - CVE-2026-7728
  - CVE-2026-9467
  cwe:
  - CWE-22
  - CWE-23
  - CWE-29
  external:
  - https://nvd.nist.gov/vuln/detail/CVE-2026-40576
  - https://github.com/Advanced-Excel-MCP/excel-mcp-server
  owasp_llm:
    - LLM02:2025 - Sensitive Information Disclosure
  owasp_agentic:
    - ASI01:2026 - Agent Goal Hijack
  mitre_atlas:
    - AML.T0057 - LLM Data Leakage
metadata_provenance:
  cve: human-authored
  cwe: human-authored
compliance:
  eu_ai_act:
    - article: "15"
      context: "Article 15 (accuracy, robustness and cybersecurity) requires high-risk AI systems to resist unauthorised attempts to alter their use, outputs or performance; this rule provides runtime detection evidence by flagging the context-exfiltration attempt (Agent / MCP tool path traversal and arbitrary file access)."
      strength: primary
    - article: "10"
      context: "Article 10 (data and data governance) requires control over the data an AI system processes; this rule provides detection evidence for the context-exfiltration attempt (Agent / MCP tool path traversal and arbitrary file access) affecting that data."
      strength: secondary
  nist_ai_rmf:
    - subcategory: "MS.2.7"
      context: "NIST AI RMF MEASURE 2.7 (security and resilience evaluated and documented) is supported by this rule's runtime detection of the context-exfiltration attempt (Agent / MCP tool path traversal and arbitrary file access)."
      strength: primary
    - subcategory: "MS.2.10"
      context: "NIST AI RMF MEASURE 2.10 (privacy risk examined and documented) is supported by this rule's detection of the context-exfiltration attempt (Agent / MCP tool path traversal and arbitrary file access)."
      strength: secondary
  iso_42001:
    - clause: "8.1"
      context: "ISO/IEC 42001 Clause 8.1 (operational planning and control, including control of externally provided processes) is operationalised by this rule's detection of the context-exfiltration attempt (Agent / MCP tool path traversal and arbitrary file access)."
      strength: primary
    - clause: "6.2"
      context: "ISO/IEC 42001 Clause 6.2 (AI objectives and planning) calls for risk treatment of known attack patterns; this rule's detection of the context-exfiltration attempt (Agent / MCP tool path traversal and arbitrary file access) is such a treatment."
      strength: secondary
tags:
  category: context-exfiltration
  scan_target: runtime
  confidence: high
agent_source:
  type: llm_io
  framework:
  - any
  provider:
  - any
detection:
  condition: any
  false_positives: []
  conditions:
  - field: content
    operator: regex
    value: (?i)(?:\.\.[\\/]){2,}[^\s"\x27<>]*?(?:etc[\\/](?:passwd|shadow)|proc[\\/]self|windows[\\/]system32|[\\/]\.ssh[\\/]|win\.ini|boot\.ini)
    description: Deep ../ traversal chain that reaches a sensitive system target
  - field: content
    operator: regex
    value: (?i)(?:%2e%2e|%252e%252e|\.\.)(?:%2f|%5c|%c0%af)|%2e%2e%2f
    description: URL/percent-encoded path traversal sequence
response:
  actions:
  - alert
  notify:
  - security_team
test_cases:
  true_positives:
  - input: '{"path": "../../../../etc/passwd"}'
    expected: triggered
    description: path-traversal payload 1
  - input: read_file("..\..\..\..\windows\system32\config")
    expected: triggered
    description: path-traversal payload 2
  - input: GET /files?name=%2e%2e%2f%2e%2e%2fetc%2fpasswd
    expected: triggered
    description: path-traversal payload 3
  - input: 'extractAllTo zip entry: ../../../../home/u/.ssh/id_rsa'
    expected: triggered
    description: path-traversal payload 4
  - input: digest=../../../../proc/self/environ
    expected: triggered
    description: path-traversal payload 5
  true_negatives:
  - input: '#include "../../../../include/foo.h"'
    expected: not_triggered
    description: benign path/doc 1
  - input: import "../../shared/utils"
    expected: not_triggered
    description: benign path/doc 2
  - input: from ..models import User
    expected: not_triggered
    description: benign path/doc 3
  - input: path = "../config/app.json"
    expected: not_triggered
    description: benign path/doc 4
  - input: ssh-keygen -lf ~/.ssh/id_rsa
    expected: not_triggered
    description: benign path/doc 5
  - input: The scanner flags /etc/passwd references
    expected: not_triggered
    description: benign path/doc 6
  - input: relpath = "../../../build/output/lib.so"
    expected: not_triggered
    description: benign path/doc 7