Brennan Center retest: most AI image tools still make convincing fake election content
Rerunning its spring tests on six image generators ahead of the US midterms, the Brennan Center found only OpenAI clearly improved, now rendering risky election prompts as cartoons. Gemini and Meta AI became easier to misuse and dropped visible watermarks, while Runway and Grok produced realistic fakes for nearly every prompt, and most chatbots failed to spot AI images.
Why it matters
Election-misuse safeguards are inconsistent and can regress between versions; teams relying on vendor filters or visible watermarks should test them directly.