Three outlets publish a similar explainer about how certain CAPTCHA challenges can function as machine-training tools. The articles focus on common CAPTCHA formats that ask users to complete image-based tasks, including selecting or identifying traffic lights. They describe these interactions as a form of “reverse centaur” activity, in which people effectively provide labels or corrections for machine-learning systems, rather than machines simply verifying humans. In this framing, the human user acts as part of the training loop: by making choices in the CAPTCHA, the system collects information about what the user perceives in the images. The outlets present this as an explanation for why such tasks work and what they can achieve beyond access control. While each publication uses the same example and core idea, all sources align on the central point that completing these CAPTCHAs contributes data that can help computers learn to recognize objects—specifically traffic lights—in future automated tasks.