[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-method-lets-ai-models-forget-one-person-not-everyone":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},8316,"new-method-lets-ai-models-forget-one-person-not-everyone","New Method Lets AI Models Forget One Person, Not Everyone","A new contrastive sparse autoencoder called SCALPEL edits AI models' internal representations to forget one person's data, not just their outputs.","A new academic method claims it can make an AI model forget a single person's data without scrambling everything else it has learned.\n\nThe system, called SCALPEL, tackles a narrow but real problem: privacy laws like the EU's GDPR sometimes require companies to erase one person's information from a trained model, not just change its behavior. The researchers found that standard interpretability tools used to locate information inside a model have a bias problem: they favor big, dominant patterns and miss small, specific details, like the record of one individual. SCALPEL is a contrastive sparse autoencoder built to hunt for those narrow, low-profile 'forget' targets instead of the loudest signal in the data. Tested on the TOFU benchmark across three open models, Qwen, Llama, and Gemma, it beat existing extraction methods like NMF and standard sparse autoencoders, and held its own against established unlearning techniques such as Gradient Difference and RMU.\n\nThis matters because most machine unlearning today works at the level of outputs: fine-tune a model until it stops saying the forbidden thing. That is not the same as removing information from the model's internals, and prior research has shown 'forgotten' data can sometimes be recovered anyway. Editing at the representation level is a stronger answer to a regulator asking whether data was actually deleted, not just suppressed.\n\nStill, this is a benchmark result on TOFU, a synthetic testbed of fictional authors, not a fix running on a production model handling a real deletion request. The gap between passing an academic unlearning test and satisfying an actual privacy audit remains wide.","[\"machine-unlearning\",\"sparse-autoencoders\",\"gdpr\",\"ai-safety\"]","2026-09-28T04:00:00.000Z","2026-09-28T23:39:28.143Z","2026-09-28T23:39:34.997Z","published",null,[],"ai",[26,27,28,29],"machine-unlearning","sparse-autoencoders","gdpr","ai-safety",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.31056",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",4900,"2026-09-28T17:44:43.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":45},"Security","security",766,"2026-09-28T15:35:23.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Policy","policy",405,"2026-09-28T17:00:51.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",269,"2026-09-28T17:42:46.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Hardware","hardware",191,"2026-09-28T15:45:00.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Consumer Tech","consumer-tech",139,"2026-09-28T17:09:47.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Dev Tools","dev-tools",87,"2026-09-28T16:11:42.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Startups","startups",80,"2026-09-28T17:50:28.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]