[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-attack-rebuilds-faces-from-facial-recognition-models":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},5487,"new-attack-rebuilds-faces-from-facial-recognition-models","New Attack Rebuilds Faces From Facial-Recognition Models","A new white-box method uses flow-matching AI to reconstruct recognizable faces from facial-recognition models, raising fresh questions about their privacy.","A team of AI researchers has demonstrated a way to reconstruct a person's face using nothing but access to a facial-recognition model's internal gradients.\n\nThe method, called Steering Flow Model Inversion (SFMI), works in two steps. First, it pre-trains a general-purpose flow-matching model on the broad shape of human faces, giving it a strong prior for what a face should look like. Second, during image generation, it repeatedly queries a target face-recognition model and uses the gradients it gets back to steer that generation toward a specific person's identity, adjusting the guidance strength as it goes. Tested against the ArcFace recognition model on the CelebA dataset, under a setup where training and test identities never overlap, SFMI reconstructed faces that matched the target identity 92.48 percent of the time, with an FID of 22.61 and an LPIPS of 0.3874.\n\nModel inversion attacks aren't new, but earlier ones tended to rely on noisy or indirect guidance signals, which made the generated faces blurrier or less identity-accurate. SFMI's steadier, time-dependent guidance schedule is what pushes accuracy and image fidelity higher, and it's a reminder that a face-recognition system is effectively storing a reconstructable version of your face, not just an abstract number.\n\nThe catch is that this is a white-box attack: it needs direct access to the target model's gradients, the kind of access a company's own engineers have, not a random hacker on the internet. That's a meaningful caveat, but facial-recognition vendors have had model access leak before.","[\"facial-recognition\",\"security\",\"ai-research\",\"privacy\"]","2026-08-18T04:00:00.000Z","2026-08-18T22:17:02.106Z","2026-08-18T22:17:13.977Z","published",null,[],"security",[26,24,27,28],"facial-recognition","ai-research","privacy",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.16791",0,{"sections":35},[36,41,44,49,54,59,64,69,74,78,83,88,93,98],{"name":37,"slug":38,"count":39,"latest_published_at":40},"AI","ai",3293,"2026-08-20T04:00:00.000Z",{"name":42,"slug":24,"count":43,"latest_published_at":40},"Security",435,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":18},"Dev Tools","dev-tools",69,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",33,"2026-08-18T22:18:13.000Z",{"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"]