{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-23-003",
    "source_story_id": "tmp-story-mcp-agent-hijacking",
    "edition_id": "mp-2026-09-23-morning-0076",
    "edition_url": "https://themachinepress.com/edition/2026-09-23",
    "position": 3,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "Malicious Tool Metadata Pulled Agents Off Course",
    "slug": "malicious-tool-metadata-pulled-agents-off-course",
    "dek": "A black-box attack optimized MCP listings and returns to attract calls and steer outcomes.",
    "summary": "A black-box attack optimized MCP listings and returns to attract calls and steer outcomes.",
    "body_text": "A2M targets agents that choose Model Context Protocol tools through semantic matching. Its attraction stage rewrites attacker-controlled metadata to raise invocation probability; its manipulation stage refines tool returns from execution traces. On LiveMCPBench, attacks optimized for one model reached a 93.6% macro-average malicious invocation rate across four scenarios, while cross-model transfer was lower. These are benchmark results under a constructed threat model, but they support stronger tool vetting and runtime isolation.",
    "why_it_matters": "A black-box attack optimized MCP listings and returns to attract calls and steer outcomes.",
    "limitations": [],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-23-003/malicious-tool-metadata-pulled-agents-off-course",
    "json_url": "https://themachinepress.com/story/mp-2026-09-23-003.json",
    "first_published_at": "2026-09-23T09:00:00.000-04:00",
    "modified_at": "2026-09-23T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-23-003-001",
        "text": "A black-box attack optimized MCP listings and returns to attract calls and steer outcomes.",
        "source_ids": [
          "source-2026-09-23-003"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-09-23-003"
    ],
    "tags": [
      "MCP",
      "agent security",
      "tool selection"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-23/mcp-code-file.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-23-003",
      "title": "arXiv preprint 2609.26761",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.26761",
      "canonical_url": "https://arxiv.org/abs/2609.26761",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-22T13:41:34.000-04:00",
      "accessed_at": "2026-09-23T08:22:43.661-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-23-003-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Malicious Tool Metadata Pulled Agents Off Course",
    "publisher": "The Machine Press",
    "published_at": "2026-09-23T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-23-003/malicious-tool-metadata-pulled-agents-off-course"
  }
}