[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-framework-helps-forgery-detectors-learn-without-forgetting":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},8989,"new-framework-helps-forgery-detectors-learn-without-forgetting","New Framework Helps Forgery Detectors Learn Without Forgetting","A new benchmark shows image forgery detectors lose accuracy on old fakes when trained on new ones, and researchers built a system to fix the tradeoff.","A team of researchers has built an AI system that keeps learning to spot new kinds of doctored photos without losing its grip on the fakes it already knows.\n\nToday's image forgery localization tools are good at flagging manipulated pixels in forgery types they were trained on, but they stumble when a new manipulation technique shows up, because in real forensic work data arrives in a stream, not all at once. No one had built a continual learning framework for this specific problem before, so the researchers made one: a benchmark with two test setups, one that swaps in data from new datasets and one that swaps in new types of forged content. Running current state-of-the-art detection and continual learning methods through it caused significant performance drops. Their fix combines a module that routes different forensic clues to different parts of an image, a prompting layer that turns those clues into structured evidence for Meta's Segment Anything Model, and a gradient-adjustment technique that protects what the model already knows about older forgery types while it adapts to new ones.\n\nThis matters because forgery detectors used by platforms, newsrooms, and forensic labs are usually trained once and left alone, while manipulation techniques, especially diffusion-based edits, keep changing. A detector that can't update without forgetting is a liability the moment a new editing tool shows up. Framing this as a continual learning problem, the same one that plagues chatbots and classifiers elsewhere in AI, is a useful reframe: forgery detection is an ongoing fight, not a one-time classification win.\n\nStill, strong numbers on a benchmark are not the same as holding up against someone actively trying to fool the detector, which is a messier problem than swapping in a new dataset.","[\"image forensics\",\"deepfakes\",\"continual learning\",\"ai research\"]","2026-10-01T04:00:00.000Z","2026-10-01T14:32:54.208Z","2026-10-01T14:32:58.215Z","published",null,[],"ai",[26,27,28,29],"image forensics","deepfakes","continual learning","ai research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.38251",0,{"sections":36},[37,40,44,49,54,59,63,68,73,77,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5487,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",809,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",429,"2026-10-01T02:26:17.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",162,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":74,"slug":75,"count":71,"latest_published_at":76},"Software","software","2026-09-30T21:41:11.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]