[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-runtime-tool-blocks-ai-agent-actions-with-unapproved-side-effects":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},8265,"runtime-tool-blocks-ai-agent-actions-with-unapproved-side-effects","Runtime Tool Blocks AI Agent Actions With Unapproved Side Effects","EffectMatch checks what an AI agent's action actually changed against what was approved, and blocks the commit if the two don't match.","A new runtime called EffectMatch stops AI agents from finishing an action if what actually changed in a system does not match what was approved.\n\nResearchers built EffectMatch to close a gap in how AI agents are supervised when they act on databases and live services instead of just generating text. An approval system might sign off on a database update, but the actual execution can also trigger side effects nobody approved, like an extra notification. EffectMatch watches the persistent changes an agent produces inside a controlled execution boundary, compares them to what was approved for that state, and only then allows the commit and any dependent steps to proceed. In tests on 206 public business tasks, it preserved every clean run and stopped every incorrect commit the researchers tried; follow-up tests, including 20-run ablations and 80 task-topology cases, showed the same result and pinpointed which safeguard failed when a given check was removed.\n\nThe gap here is real and easy to miss. Most agent guardrails check the request an agent makes or log what happened afterward - neither actually verifies the result before the system moves on. As agents get write access to production databases, CRMs, and messaging tools, that blind spot is where an approved-looking action quietly does something nobody signed off on.\n\nEffectMatch is a research prototype tested on curated benchmark tasks, not a drop-in product, and 206 business tasks is a tidy sandbox compared to the mess of real production systems.","[\"ai-agents\",\"ai-safety\",\"dev-tools\"]","2026-09-28T04:00:00.000Z","2026-09-28T20:24:35.646Z","2026-09-28T20:24:42.318Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The headline says the tool 'flags' unapproved side effects, but the body describes it actively blocking the commit and dependent steps—rewrite the headline\u002Fdek to reflect that it blocks, not merely flags.","resolved","ai",[32,33,34],"ai-agents","ai-safety","dev-tools",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.31301",0,{"sections":41},[42,46,51,56,61,66,71,76,81,85,90,95,100,105],{"name":43,"slug":30,"count":44,"latest_published_at":45},"AI",4900,"2026-09-28T17:44:43.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Security","security",766,"2026-09-28T15:35:23.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Policy","policy",405,"2026-09-28T17:00:51.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Deals","deals",269,"2026-09-28T17:42:46.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",191,"2026-09-28T15:45:00.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Consumer Tech","consumer-tech",139,"2026-09-28T17:09:47.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":82,"slug":34,"count":83,"latest_published_at":84},"Dev Tools",87,"2026-09-28T16:11:42.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Startups","startups",80,"2026-09-28T17:50:28.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":106,"slug":107,"count":108,"latest_published_at":109},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]