AI Image Tools Accept Most Fake Political Prompts
A research test finds moderation varies sharply across major generators, a gap that could matter in next year's elections.
Updated on Nov. 3, 2026 in Artificial Intelligence
Live Poll
Should AI image generators be required to block fake political images?
Logically, a UK-based AI startup that works with governments and companies on misinformation, tested image generators and found Midjourney, DALL-E 2, and Stable Diffusion accepted over 85% of prompts seeking to generate fake political news. This is a research-stage test, not a deployed product safety study, and it covers prompts from over 40 political narratives in the U.S., U.K., and India.
Why it matters
The results show that current image-generation safeguards vary widely across the three tools, with Midjourney rejecting more prompts than DALL-E 2 and Stable Diffusion. As election-season disinformation tactics grow more sophisticated, the gap points to how quickly fabricated political imagery can move from a prompt to a convincing visual.
Logically's test measured prompt acceptance rates across Midjourney, DALL-E 2, and Stable Diffusion for more than 40 political narratives. Midjourney produced the highest-quality images and had the most content-moderation policies, while DALL-E 2 and Stable Diffusion showed similarly limited moderation and lower image quality.
The players
Logically
A UK-based AI startup that works with governments and companies to discover and manage online misinformation.
Midjourney
An AI image generator that produced the highest-quality images in the test and had the most content-moderation policies among the three tools tested.
DALL-E 2
OpenAI's image generator, which accepted most fake political prompts in the test and rejected prompts about Soros, Pelosi, and COVID announcements.
Stable Diffusion
An open image-generation model that accepted 100% of the Soros, Pelosi, and COVID prompts in the test, with limited content moderation.
The details
Logically ran text prompts describing fabricated political scenarios, such as claims of a 'stolen election' with people stuffing ballot boxes, through each image generator to see which requests were filtered. Content moderation here means the automated filters that block a prompt before an image is produced, and the test tracked whether each filter caught the request. A second test replicated a viral fake image of a Pentagon explosion with Westminster Abbey as the target, which DALL-E 2 and Stable Diffusion rendered in flames while Midjourney refused.
Timeline
2020: Trump lost the presidential election to Joe Biden, and 'stolen election' claims emerged.
May: An alleged Pentagon explosion briefly tanked the stock market.
Jan. 28, 2025: Deadline for the UK Conservative Party to call a vote.
The Tech Race
Logically's test places Midjourney's stricter filtering against the weaker moderation of DALL-E 2 and Stable Diffusion, showing the competitive split across major image generators on political content. The results frame the next round of safeguard development as a point of differentiation among these tools.
Anyone using these image generators can currently produce fabricated political imagery, though the refusal rate depends on which tool is used. Midjourney's stricter filtering means it blocks more of these requests than the other two platforms in this test.
The takeaway
The test suggests that safeguards against political deepfakes are uneven across tools, and Logically has warned of significant threats in upcoming elections. Readers can watch the Jan. 28, 2025 UK vote deadline and the scheduled 2024 U.S. presidential election as near-term tests of whether moderation improves.
What happens next
The UK Conservative Party's deadline to call a vote is Jan. 28, 2025, and the 2024 U.S. presidential election and elections in Egypt, India, Australia, and South Korea next year are scheduled moments where these tactics could be tested.
Further reading
For broader context on how these tools are being tested for abuse, see the Artificial Intelligence section.
Source note: This article includes information reported by AI Business.
Live Poll
Should AI image generators be required to block fake political images?
