[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-forget-me-not-tests-have-a-blind-spot":10,"sections":34},{"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":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3775,"ai-forget-me-not-tests-have-a-blind-spot","AI Forget-Me-Not Tests Have a Blind Spot","A new benchmark finds that erasing private individuals from AI models tends to collateral-damage the public information pictured alongside them.","Researchers want to know if AI forgetting tools actually work — and the answer is: sort of, but with collateral damage.\n\nA new benchmark called PPE-Bench tests whether multimodal large language models can selectively forget private individuals without wiping out surrounding public context. The problem it exposes is specific: when a private person appears in a photo next to a celebrity or in front of a famous landmark, current unlearning methods tend to degrade the model's knowledge of that public information too. Prior benchmarks had sidestepped this by using clean, isolated images — one person, no background complexity — which bears little resemblance to the messy photos that actually end up in training data. PPE-Bench introduces images where the target to be forgotten is visually entangled with public figures or recognizable places, and it pairs the benchmark with two new methods aimed at better preserving that collateral context. Experiments show that existing unlearning approaches do reduce private information leakage, but they frequently cause substantial harm to adjacent public knowledge.\n\nThis matters because machine unlearning is increasingly pitched as the tidy fix for AI privacy — a surgical option that avoids the cost and disruption of retraining a model from scratch. If the surgery routinely nicks what it wasn't supposed to touch, that pitch gets harder to sustain. Regulators in Europe already treat the right to erasure as a legal obligation, not a nice-to-have, and the gap between compliance theater and genuine forgetting is exactly the kind of gap a benchmark like this is built to expose.\n\nFor now, unlearning remains a promising idea with a messy implementation record — and PPE-Bench just made it harder to hide that.","[\"ai\",\"privacy\",\"machine-learning\",\"benchmarks\"]","2026-07-07T04:00:00.000Z","2026-07-07T08:14:19.131Z","2026-07-07T08:14:22.105Z","published",null,[],"ai",[24,26,27,28],"privacy","machine-learning","benchmarks",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.02897",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]