Surrounded Planets (AI-agent detection trap)

Surrounded Planets is a detection item, not a research paradigm: it separates genuine human respondents from vision-enabled AI agents completing surveys autonomously. It exploits cross-modal binding with occlusion — vision-language models struggle to bind visual features (here, the surrounding shapes) to the correct object (the planet) when the objects overlap — following the Cognitive Trap Framework of Affonso (2026, Journal of Consumer Research).

Participants see six coloured planets (orange, purple, blue, red, green, gray), each ringed by several small geometric shapes that partly overlap the planet. Five planets are surrounded by 5 shapes; exactly one is surrounded by only 4. The task is to find that planet and name its colour. A human counts the shapes around each planet and finds the odd one almost immediately. A vision-language model, unable to reliably bind each shape to a single planet under occlusion, tends to mis-count and pick the wrong colour.

In the reference data (Affonso 2026), this trap produced ~55% agent-failure with ~96% human pass. It is the least uniform trap across model families and is a maturing capability (some models are improving on it), so it is best used as a supporting item alongside the more durable illusion and spatiotemporal traps.

Deploy as part of a battery, not alone. The recommended screening rule is to flag respondents who fail 3 or more of 5 such traps. Which colour is the odd (4-shape) planet, the colour layout, and the shapes are drawn at random per session from a pool of validated figures, so the correct answer cannot be memorised. See the Deployment notes below.