Уровень 0 · материалов: 6
В кластер входят документы об уязвимостях и способах обмана конкретных технических систем контроля или мониторинга, но не входят общие методы обеспечения приватности или обсуждение правовых норм.
Общие признаки: технические уязвимости, обход систем слежения, состязательные атаки, ошибки автоматизированных систем, манипуляция данными
Группа выше: Утечки через ошибки доступа и предсказуемые идентификаторы
Смысл: The main idea is that scientific knowledge and technical documentation can be used to successfully challenge the accuracy of automated law enforcement tools when those tools are operated outside their specified technical parameters.
A Kazan physicist successfully overturned a speeding fine by proving in court that the GIBDD's radar was improperly installed, leading to inaccurate readings based on the Doppler effect.
Смысл: The main idea is the exploration of adversarial fashion as a means of 'hardware hacking' to protect privacy against AI surveillance, specifically demonstrating how visual patterns can deceive the YOLO detection algorithm.
Rachel Didedro created the 'Manifesto' clothing line to trick AI surveillance systems like YOLO into misidentifying humans as animals to protect user privacy.
Смысл: The text describes the testing of the FAST system by the US government, which aims to predict criminal intent using biometric sensors and AI, while noting the skepticism of experts regarding its claimed high accuracy.
The US is testing the FAST AI system to detect criminal intent via remote biometric sensors, claiming up to 80% accuracy despite expert skepticism.
Смысл: The text explores the tension between administrative efforts to enforce school attendance through biometric technology and the practical vulnerability of such systems to deception.
Schools in Washington County, Florida, have introduced fingerprint scanners to prevent attendance fraud, though the author questions their reliability based on 'Mythbusters' tests.
Смысл: The main idea is that electronic gradebooks are currently unreliable due to a combination of poor user security habits (human factor) and critical software vulnerabilities (lack of server-side validation), which together allow for the easy manipulation of academic records.
Electronic gradebooks are highly vulnerable to grade tampering because of poor teacher digital hygiene and critical flaws in server-side data validation.
Смысл: The main idea is the transition of adversarial attacks from fragile 2D digital images to robust 3D physical objects, highlighting a critical security vulnerability in computer vision systems used in the real world.
Researchers have created the first 3D-printed adversarial objects that consistently trick neural networks into misidentifying them from any angle, posing potential risks for technologies like self-driving cars.