[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-openai-shows-adversarial-images-can-fool-vision-models-from-any-angle":10},{"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":22,"tags":34,"sources":38,"feedback":42,"feedback_at":22,"cost_usd":42,"total_tokens":42},1186,"openai-shows-adversarial-images-can-fool-vision-models-from-any-angle","OpenAI shows adversarial images can fool vision models from any angle","OpenAI released a set of images that consistently mislead neural nets, undermining claims that multi‑view sensing protects self‑driving cars.","OpenAI unveiled a collection of adversarial pictures that trick image classifiers even when the viewer changes scale or perspective.\n\nThe team generated images that, across a range of zoom levels and viewing angles, are still misidentified by standard neural network classifiers. The work directly counters a recent claim that autonomous‑vehicle cameras, which capture scenes from many viewpoints, would be difficult to deceive.\n\nIf self‑driving systems rely on the same visual models, the finding suggests a wider attack surface than previously thought. It gives researchers a concrete test set for hardening perception pipelines, and forces car makers to consider defenses beyond simply adding more cameras.\n\nThe demo was posted on July 17, 2017, and includes hundreds of test images. While OpenAI offered no timeline for broader release, the result adds a data point to a growing body of work showing that visual AI remains vulnerable to carefully crafted inputs, even in real‑world‑like conditions.","[\"machine-learning\",\"adversarial-attacks\",\"autonomous-vehicles\"]","2017-07-17T07:00:00.000Z","2026-06-16T16:21:50.101Z","2026-06-16T16:21:52.920Z","published",null,[24,30],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Add a concise concluding paragraph that summarises the implications and why readers should care.","resolved",{"id":31,"reviewer":26,"round":32,"reason":33,"status":29},"editor-r2",2,"Add a concise concluding paragraph that summarises the implications and why readers should care, and include any concrete details (date of release, size of image set, any quoted OpenAI statements) to strengthen the lead and context.",[35,36,37],"machine-learning","adversarial-attacks","autonomous-vehicles",[39],{"name":40,"url":41},"OpenAI","https:\u002F\u002Fopenai.com\u002Findex\u002Frobust-adversarial-inputs",0]