A designer made a shirt that confuses AI cameras — a reminder for business

ER10 covered an unusual project: a designer created a shirt whose print «blinds» computer-vision systems — algorithms stop correctly recognizing the person in frame. It continues the adversarial-fashion line, where patterns are engineered to exploit neural networks' weak spots: the model sees millions of confusing features in the fabric and loses confidence.

The project is artistic, but it vividly demonstrates what engineers have long known: every recognition system has blind spots that can be computed and exploited. While AI cameras count visitors and watch conveyor quality, the price of such deception is small. But the more decisions hang on «the camera will see everything» — access control, violation logging, security — the more attractive fooling it becomes.

The takeaway for business is not to abandon computer vision but to build control in echelons: critical points should rest on at least two independent signals — camera plus sensor, camera plus account, camera plus human. AI in a control system is a strong layer — but a layer, not the only wall.