AI images are getting harder to spot as fake signs fade

AI-generated images and deepfake videos are now common enough to show up in scams, identity theft, propaganda and election-related misinformation, while the old visual giveaways are fading fast. Distorted fingers, unnatural eyes, smudgy chins, weird hands and dodgy numbers once helped people spot fakes; newer models are making those clues far less reliable.

That shift has pushed verification out of the realm of quick visual judgment and into source checking. Seattle-based TrueMedia launched a quiz on June 20, 2024 to test how well people can identify manipulated images and videos, and the MIT Media Lab’s DetectFakes experiment asks users to see how accurately they can identify AI-generated images. Northwestern University material tied to DetectFakes makes the same point in sharper form: spotting synthetic faces is a measurable challenge, not a simple instinct.

The scale of the problem became easier to see after a fake image of Pope Francis in a white puffer jacket spread widely and helped bring synthetic imagery into mainstream awareness. Once that image circulated, the question for journalists and readers was no longer whether AI could imitate a photo, but how often a convincing fake could move before being checked.

That is where the recent guidance has shifted. The Reuters Institute for the Study of Journalism tested publicly accessible AI detectors in February 2024 and examined where they fail, showing that software can miss manipulated images even when users expect it to be authoritative. Columbia Journalism Review followed with a 2025 guide for journalists on deepfake detection technology, underscoring that the tools themselves require caution.

A 2025 Springer Nature review also pointed to advances in deepfake detection techniques and future prospects, but the broader pattern is clear: detection is improving, yet generative tools are advancing quickly enough that both machines and human judgment have limits. For elections, the risk is obvious, because a polished fake can be built to resemble campaign imagery or post-vote scenes. For scams, a synthetic face or video can support identity theft or fraud. For breaking news, the safest test is no longer whether an image looks odd, but whether its provenance holds up, whether the original source is identifiable, and whether other evidence matches what the picture claims to show.