{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreibfla676b5ft5xrqk6ewhb3xwezmgrvvfpyjap5lqxlbrh5utxj4q",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mmaxrdg42w22"
  },
  "path": "/t/physical-ai-safety-ownership-and-execution-boundaries/175776#post_13",
  "publishedAt": "2026-05-20T01:40:31.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "#12 identifies what is structurally missing.\n\nThis comment focuses on how to complete it.\n\nTwo types of context must be declared and delivered:\n\n**Manufacturer context** is a structural issue.\nIf it was declared but not delivered to the LLM, the failure is not a model failure.\nIt is a context-delivery failure.\n\n**User context** is a methodological issue.\nThe common approach is to add more data to improve inference.\nThis framework takes a different position: if intent is incomplete, ask.\n\nAsking serves three roles at once:\n**it completes context, obtains user approval before execution, and reduces inference cost.**\n\nThe goal is not to maximize context size.\nThe goal is to make context accurate enough for responsible execution.\n\nIf manufacturer context is missing, the agent should not proceed as if nothing is missing.\nIf user context is incomplete, the agent should ask.",
  "title": "Physical AI Safety: Ownership and Execution Boundaries"
}