[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-research-flags-gaps-between-ai-agent-approvals-and-actual-work":10,"sections":40},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},9756,"research-flags-gaps-between-ai-agent-approvals-and-actual-work","Research Flags Gaps Between AI Agent Approvals and Actual Work","A new paper finds that automated approvals in AI agent marketplaces often don't prove the delivered work met the task, not that the work was bad.","A new academic study says the real risk in AI agent marketplaces isn't the AI. It's the handoff between \"approved\" and \"actually correct.\"\n\nResearchers systematized security and economic requirements for decentralized agent tasks into 17 property families spanning six stages, from task assignment through verification and settlement. They examined 12 existing systems and standards, five reusable mechanism families, and four classical baselines, then introduced a new yardstick called \"guarantee closure\" to check whether a safeguard established early in a workflow actually holds by the time a later decision depends on it. To test it, they ran 840 matched executions and an exhaustive 11,648-case check over a finite task domain.\n\nThe results matter because escrow and approval mechanisms are supposed to be the trust layer that lets autonomous agents pay each other without a human checking the work. The study finds that an authorized approval often provides little evidence that delivered work actually met the task, and that conforming work can still go unaccepted while valid evidence gets ignored elsewhere in the pipeline. That is a documented gap between verification and settlement, not proof that any specific payout went to substandard work.\n\nSmart contracts have always been good at enforcing what they can see, like whether a signature exists, and bad at judging what they cannot, like whether the work behind that signature was any good. This paper just puts a number on how often that blind spot shows up.","[\"ai\",\"security\",\"agent-economies\",\"research\"]","2026-10-02T04:00:00.000Z","2026-10-03T09:12:49.736Z","2026-10-03T09:12:55.832Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The dek claims payment was approved for work that 'does not actually meet the task,' but the source and body only support a narrower claim: approvals didn't prove conformance (an evidentiary gap), not that the delivered work was confirmed to be substandard — rewrite the dek to match that hedged, evidence-based finding.","resolved","security",[32,30,33,34],"ai","agent-economies","research",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.01756",0,{"sections":41},[42,45,48,52,57,61,65,70,75,80,85,90,95,100],{"name":43,"slug":32,"count":44,"latest_published_at":18},"AI",6041,{"name":46,"slug":30,"count":47,"latest_published_at":18},"Security",848,{"name":49,"slug":50,"count":51,"latest_published_at":18},"Policy","policy",439,{"name":53,"slug":54,"count":55,"latest_published_at":56},"Deals","deals",317,"2026-10-01T22:00:00.000Z",{"name":58,"slug":59,"count":60,"latest_published_at":18},"Hardware","hardware",199,{"name":62,"slug":63,"count":64,"latest_published_at":18},"Science","science",176,{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",155,"2026-10-01T19:54:10.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Dev Tools","dev-tools",96,"2026-10-01T16:57:03.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Software","software",93,"2026-09-30T21:41:11.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Startups","startups",90,"2026-10-01T21:55:22.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]