AI Is Going Just Great
← Timeline
·1w agoConcerningModerategoogle

Google's SynthID Detector Flagged Real Politicians' Photos as AI-Generated

Published · updated · curated by AI Is Going Just Great

Source: fullfact.org ↗

"In very rare cases the tool can falsely trigger on non-watermarked content."

Google fast-tracked an update to its public SynthID Detector tool after Full Fact found it classifying five apparently genuine images as "made or edited with Google AI." The images, posted by UK politicians including an SNP councillor, a Liberal Democrat MP's guide dog account, and two Labour MPs, showed no visible signs of manipulation — and each subject denied using AI. One flagged image was a campaign photo; another was a picture of a dog on a patch of grass, where the tool's heatmap helpfully localized the "watermark" to the grass itself.

After Full Fact contacted Google with its findings, the company rolled out an update on 15 October that reclassified the five images as "Unsure" rather than definitively AI-generated. A sixth image, posted by Sinn Féin MP Órfhlaith Begley, was still flagged as "made or edited with Google AI" even after the update. Google acknowledged that using the tool "at scale without prior suspicion of AI editing" raises the likelihood of false positives — a notable caveat for a tool marketed to the public as an aid for spotting AI-generated misinformation.