[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-deepfake-detectors-are-easier-to-fool-than-benchmarks-suggest":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},6347,"deepfake-detectors-are-easier-to-fool-than-benchmarks-suggest","Deepfake Detectors Are Easier to Fool Than Benchmarks Suggest","A new study shows deepfake detectors look robust in isolated tests but become far easier to fool once attacks are pooled across multiple source models.","Deepfake detectors that look robust in isolation crumble once attackers pool tricks from several models at once, new research shows.\n\nResearchers tested adversarial transfer attacks across 60 deepfake detectors built from six backbones, two pretraining setups, and five training-data configurations. They used two attack methods, AutoAttack and a Carlini-Wagner variant paired with Expectation over Transformation, generating adversarial images on one surrogate detector and firing them at a different target. Transfer worked best when the source and target detector shared the same backbone, architecture family, pretraining regime, or training data. Which factor mattered most depended on the attack: exact backbone matches drove success under AutoAttack, while shared pretraining and training data mattered more under the Carlini-Wagner method.\n\nAveraged across single attack sources, success rates looked modest: 7.21 percent for AutoAttack and 19.52 percent for the Carlini-Wagner approach. But combining both attacks across multiple source models pushed the mean success rate to 64.48 percent, even after excluding the easiest matched-backbone and matched-data cases. That gap matters because most detector robustness claims rest on single-source testing, which this work suggests badly understates how exploitable a deployed system really is.\n\nThe team released 240,000 adversarial images and the full pairwise transfer results, giving other researchers a shared benchmark instead of another isolated robustness claim to take on faith.","[\"deepfake-detection\",\"adversarial-attacks\",\"ai-security\",\"robustness-research\"]","2026-09-11T04:00:00.000Z","2026-09-11T07:36:56.359Z","2026-09-11T07:37:08.268Z","published",null,[],"security",[26,27,28,29],"deepfake-detection","adversarial-attacks","ai-security","robustness-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.10002",0,{"sections":36},[37,41,44,48,53,58,63,66,71,75,80,85,90,95],{"name":38,"slug":39,"count":40,"latest_published_at":18},"AI","ai",3522,{"name":42,"slug":24,"count":43,"latest_published_at":18},"Security",637,{"name":45,"slug":46,"count":47,"latest_published_at":18},"Policy","policy",338,{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",153,"2026-09-09T15:12:32.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":62},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":64,"slug":65,"count":61,"latest_published_at":18},"Science","science",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":18},"Dev Tools","dev-tools",70,{"name":76,"slug":77,"count":78,"latest_published_at":79},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]