[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-safety-framework-cuts-attack-success-rate-to-31":10,"sections":35},{"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":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},11052,"ai-safety-framework-cuts-attack-success-rate-to-31","AI Safety Framework Cuts Attack Success Rate to 31%","A new framework repeatedly red-teams and patches AI models, cutting jailbreak success rates from 86% to 31%, below a top frontier model's 57%.","A new AI safety framework trains models to patch their own vulnerabilities automatically, and it is beating the industry's best defense numbers.\n\nThe framework, called ReSI (Recursive Safety Improvement), runs models through repeated rounds of automated red-teaming, then builds training updates that fix whatever weaknesses turn up. Each round only keeps the update with the biggest safety gain among those that pass a check for not degrading the model's general capabilities. Researchers tested the approach on four models, a mix of dense and mixture-of-experts architectures. Across those four, ReSI cut the average success rate of X-Teaming attacks - a method for probing models with adversarial prompts - from 86.01% to 31.45%, a result the researchers say beats GPT-5.6-Luna's reported 56.69% on the same benchmark.\n\nAI labs keep shipping new checkpoints, and each one needs its own safety pass, a process that currently leans on manual red-teaming and ad-hoc patches. Automating that cycle - find the hole, patch it, verify nothing else broke - is the kind of unglamorous infrastructure work that decides whether safety keeps pace with capability, not whether a chatbot can write poetry.\n\nThe paper frames this as recursive self-improvement applied to safety rather than raw capability, a tidy reframe, but beating one named frontier model on one attack benchmark is a narrower claim than solving alignment at scale.","[\"ai safety\",\"red-teaming\",\"alignment\",\"research\"]","2026-10-09T04:00:00.000Z","2026-10-10T03:53:40.734Z","2026-10-10T03:53:46.485Z","published",null,[],"ai",[26,27,28,29],"ai safety","red-teaming","alignment","research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.12233",0,{"sections":36},[37,40,44,49,54,58,62,67,72,77,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",6784,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",932,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",486,"2026-10-08T22:40:11.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",474,"2026-10-08T22:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":18},"Hardware","hardware",232,{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",193,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",181,"2026-10-08T23:26:35.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Startups","startups",117,"2026-10-08T16:45:00.000Z",{"name":73,"slug":74,"count":75,"latest_published_at":76},"Software","software",114,"2026-10-08T17:57:01.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":18},"Dev Tools","dev-tools",106,{"name":82,"slug":83,"count":84,"latest_published_at":85},"General","general",66,"2026-10-09T04:46:11.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Gaming","gaming",58,"2026-10-08T20:08:45.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",34,"2026-10-08T14:00:22.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]