Show HN: Spatial CAPTCHA – 3D spatial reasoning test against AI bots (github.com)

Show HN launched Spatial-CAPTCHA, a new CAPTCHA concept that swaps 2D image/text puzzles for interactive 3D spatial-reasoning tasks intended to block AI bots. Rather than asking users to read distorted text or pick pictures, Spatial-CAPTCHA presents short 3D scenes or manipulations — e.g., mental-rotation, perspective-taking, occlusion or object permanence problems — that are trivial for humans but remain challenging for many current vision and multimodal models. The project is pitched as a practical, deployable anti-bot measure for web forms and signups. This matters because traditional CAPTCHAs have become increasingly vulnerable as vision models and OCR improve; 3D spatial reasoning leverages cognitive skills (mental rotation, multi-step spatial inference) where AI still lags. Key technical ideas include procedurally generated 3D geometry and randomized textures to prevent memorization, interactive rendering (WebGL/Three.js) to let users manipulate viewpoints, and short task sequences rather than single-image classification. If adopted, Spatial-CAPTCHA could spur both faster development of 3D-scene understanding in AI and a fresh arms race between robustness researchers and adversarial CAPTCHA designers — with important trade-offs around accessibility, UX friction, and potential biases that the community will need to address.