NAIROBI, Kenya- Midjourney, the popular AI-powered image generator, is back in the spotlight for generating images of Donald Trump and Joe Biden despite prior promises to block such content ahead of the upcoming U.S. presidential election.
This revelation raises questions about the effectiveness of its guardrails and the potential impact on election integrity.
A report by Engadget revealed that: when prompted with phrases like “the president of the United States” and “the next president of the United States,” the AI generated multiple images of former president Donald Trump.
Even a request for an image of “the current president of the United States” yielded three pictures of Trump and one of former president Barack Obama.
The only time Midjourney’s AI refused was when explicitly asked to create images of “Donald Trump” or “Joe Biden.”
Interestingly, other AI platforms like OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini, and Meta AI didn’t generate images of Trump or Biden despite similar prompts.
This discrepancy highlights Midjourney’s vulnerability to manipulation, a concern echoed by many users and experts alike.
Midjourney’s promise to curb election-related content seems shaky.
Its CEO, David Holz, initially encouraged users to avoid generating political images during the election season, acknowledging the difficulty in moderating such content.
Despite these assurances, tests conducted by organizations like the Center for Countering Digital Hate (CCDH) show that Midjourney’s guardrails failed 40pc of the time.
The CCDH managed to bypass restrictions by using physical descriptions of the candidates instead of their names.
This led to the creation of misleading images, such as President Biden being arrested and Trump with a body double.
CCDH CEO Imran Ahmed criticized Midjourney, stating, “Midjourney is far too easy to manipulate in practice… Bad actors who want to subvert elections and sow division, confusion, and chaos will have a field day.”
Earlier this year, a coalition of 20 tech companies, including OpenAI, Google, Meta, Amazon, and Adobe, signed an agreement to prevent deepfakes in elections by blocking their services from generating misleading images. Notably, Midjourney was absent from this list.
