CCDH: Midjourney, DreamStudio, ChatGPT Plus, and Microsoft Image Creator created election disinformation in 41% of tests; Midjourney was most likely to do so
New York CNN — — Leading artificial intelligence image generators can be manipulated into creating misleading election-related images …
Context & Ripple Effects
This audit puts election misuse at the image-generation layer: safeguards at four consumer AI services did not reliably block prompts for misleading political imagery. It extends a longer problem in which platforms have struggled to identify manipulated visual media, as a prior analysis of detection systems found.
The later coverage suggests the risk is not confined to image tools: chatbots also repeated misinformation, while ChatGPT deployed election-period refusals and news redirects around voting. That makes model-level controls, not just downstream moderation, a central operational issue.
First-order effects
- Midjourney, DreamStudio, ChatGPT Plus and Microsoft Image Creator face immediate scrutiny of their election-safety guardrails; Midjourney’s higher failure rate makes it the clearest target for remediation.
- People seeking to create deceptive election imagery have evidence that ordinary consumer generators can be probed for bypasses, raising the verification burden for audiences and election-information providers.
Second-order effects
- Social platforms, newsrooms and fact-checkers must expect a larger and cheaper supply of plausible political visuals, increasing the need to assess provenance and context rather than image realism alone.
- Competing AI providers are pressured to test and publish stronger prompt restrictions, especially as later reporting found deepfake-related Community Notes rose alongside image-model releases.
Third-order effects
- If repeated audits continue to find bypasses, election integrity will increasingly depend on a synthetic-media control plane spanning model access rules, provenance signals and distribution-platform responses—not any single model’s policy.
- The pattern supports more formal dual-use AI governance for broadly available creative tools, though the corpus does not establish which technical or regulatory approach will prove effective.
The trend: Consumer generative AI is shifting election-disinformation risk from specialized manipulation toward scalable, model-enabled content production, forcing safety controls upstream into the tools themselves.