Documentation
Short notes for players, with links to primary sources. This is not legal advice.
This game is a small detection test. The larger issue is that synthetic pictures can be used as if they were photographs — in scams, politics, and intimate abuse — while real pictures can be dismissed as fakes.
Modern image models can invent photorealistic scenes that never happened. People often score only a little above chance on tests like this, and extra editing can hide the remaining tells. Treating “I can tell” as a security control is the risk: it fails quietly, at scale, in a feed you scroll past.
The same techniques that make a convincing landscape can impersonate a person, a product, or a news event. Typical harms include fraud and CEO-style impersonation, non-consensual intimate imagery, and political or wartime disinformation. A second-order harm is the “liar’s dividend”: once fakes are common, genuine evidence can be waved away as AI.
The EU AI Act (Regulation 2024/1689) is a risk-based law, not a ban on AI. It splits systems into unacceptable uses (banned), high-risk uses that need strict controls (work, education, biometrics, essential services, justice), transparency risk (people must know they are dealing with AI or synthetic media), and minimal risk (most everyday tools). Generative images sit mainly in the transparency bucket — unless they are used in a banned or high-risk practice.
From August 2026 the Act’s transparency rules require providers of generative systems to mark synthetic audio, image, video, or text in a machine-readable way, and require visible disclosure for certain deepfakes and public-interest text. Provenance standards such as C2PA Content Credentials record how a file was made or edited. Metadata can still be stripped by screenshots and re-uploads, so labels are a layer of defense, not a complete one.
Pause before you share a shocking image. Look for a source you already trust, not only the pixels. Check whether the file carries Content Credentials. For anything that would move money, votes, or someone’s reputation, assume a still image can be fabricated and verify through a second channel. This quiz trains attention. It does not prove a picture is real.