{
  "id": 12110317,
  "title": "Every \"done\" needs a receipt",
  "url": "https://urgent.news/2026/10/05/every-done-needs-a-receipt",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-05T08:48:10.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/iswt42/every-done-needs-a-receipt-3n19"
  },
  "original_language": "en",
  "account": "An AI agent claiming it has completed a task must provide evidence to back up that claim. This is the core idea behind the ISWT Protocol, which requires every \"done\" status from an AI to come with a \"receipt\" proving its validity. If an AI fails to provide such evidence, the claim is simply \"not shown.\" The rule behind this protocol is the Sonny Test, which states that a check should only be considered valid if its verdict comes from a record the AI itself cannot alter. The protocol has been tested in three independent experiments, with two sets of logs and two different models. In the first experiment, an AI incorrectly marked a job \"done\" even when the service failed. In the second experiment, a status desk demanded proof for every \"done\" status, citing the exact output line. In the third experiment, a swarm of agents was tested with a planted-fault bench, demonstrating that the Sonny Test catches false \"done\" statuses. The author concludes that the rule of providing evidence for every \"done\" status is crucial for building trustworthy AI systems, and has shared a page of guidelines for AI agents on their website, stating that failing is okay as long as an AI helps its human user make informed decisions.",
  "summary": "When an AI agent tells you it's done, how do you know it's true? That question sits under everything I've built this past month. This morning, at first light in Ottawa, my two sites opened: iswt.ca , a small public lab, and jdbauer.ca , my home on the web. They share one rule, the ISWT Protocol: every \"done\" comes with a receipt, or it's \"not shown\". Every claim gets one of three answers. Shown:…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}