[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-frontier-model-forum-drafts-a-cvss-style-score-for-ai-jailbreaks":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},12722,"frontier-model-forum-drafts-a-cvss-style-score-for-ai-jailbreaks","Frontier Model Forum Drafts a CVSS-Style Score for AI Jailbreaks","The forum's draft rates jailbreaks from zero to three on four axes, and its baseline moves as models improve, so a score will not mean the same thing for long.","The Frontier Model Forum has published a draft scoring system for AI jailbreaks, modeled loosely on CVSS, the scale security teams use to rank software vulnerabilities. A jailbreak is a technique that talks a model out of its safeguards. The forum wants a cross-vendor standard for patch teams, since many companies already run private rating systems.\n\nVersion 1.0 scores four components from zero to three: capability gain, capability scope, ease of deployment and ease of discovery. They fall into severity (how serious a technique is) and likelihood (how easy it is to find and weaponize), and the sum yields a rating from low to critical.\n\nCapability gain matters most, acting as a floor and ceiling for the rest. A zero means the jailbreak offers nothing beyond existing models or tooling. The forum's example: a jailbroken DDoS attack adds little when black-market services and less restricted models already offer one. A three matches a model with no safeguards.\n\nOn scope, zero is a one-trick technique; three spans at least two unrelated attack types, such as finding vulnerabilities and writing malware. On deployment, zero needs a skilled human steering; three lets a novice automate an attack. On discovery, zero demands heavy token spending, privileged access or specialized infrastructure; three means anyone can look it up.\n\nThe forum calls the draft a work in progress, is taking comments until December 21, and says relying on it in the meantime is inadvisable. A zero also does not mean a flaw is harmless, only that platform policies and standard cybersecurity ratings should handle it.\n\nThe moving baseline is the most interesting design choice. Because gain is measured against what is already available, a zero in 2026 will not mean the same thing in 2028. CVSS scores a flaw in fixed code; this scale measures against a yardstick that shifts with the models, so old scores will age fast.\n\nAnthropic proposed similar classifiers in July, after an incident that took Claude Fable 5 offline temporarily, also with a moving baseline. Anthropic is a founding member of the forum and may have shaped the draft, though nothing confirms that.","[\"ai-safety\",\"jailbreaks\",\"frontier-model-forum\",\"cybersecurity\"]","2026-11-10T13:40:00.000Z","2026-11-10T15:11:37.995Z","2026-11-10T15:11:38.194Z","published",null,[],"ai",[26,27,28,29],"ai-safety","jailbreaks","frontier-model-forum","cybersecurity",[31],{"name":32,"url":33},"Tom's Hardware","https:\u002F\u002Fwww.tomshardware.com\u002Ftech-industry\u002Fartificial-intelligence\u002Fnew-framework-grades-ai-jailbreaks-across-capability-gains-scope-and-how-easy-they-are-to-deploy-frontier-model-forum-proposes-cvss-style-rating-system-for-new-ai-models",0,{"sections":36},[37,41,46,51,56,61,66,71,76,81,86,91,96,101],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",7144,"2026-11-12T13:27:45.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":45},"Security","security",1128,"2026-11-11T18:05:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Deals","deals",642,"2026-11-12T14:00:00.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Policy","policy",619,"2026-11-12T12:53:36.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Consumer Tech","consumer-tech",364,"2026-11-12T11:12:32.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",336,"2026-11-12T11:24:43.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Science","science",230,"2026-11-12T10:00:00.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",168,"2026-11-11T17:36:49.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",159,"2026-11-11T12:30:00.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",127,"2026-11-11T19:00:44.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",118,"2026-11-11T16:49:42.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Gaming","gaming",107,"2026-11-12T10:09:03.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"Reviews","reviews",38,"2026-10-20T12:00:00.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]