DM Monitoring
WASHINGTON: Tests on a leading AI tool allowed the creation of deceptive and incriminating images of President Joe Biden and Donald Trump, a watchdog said Wednesday, despite pledges to block fake photos of the presidential contenders ahead of elections in November.
Disinformation researchers fear rampant misuse of AI-powered applications in a year of major elections around the world, thanks to the proliferation of cheap online tools that are easy to use and lack sufficient guardrails.
The non-profit Centre for Countering Digital Hate (CCDH) said it tested two programs that can generate images based on text prompts — Midjourney and ChatGPT, from Microsoft-backed OpenAI.
“Midjourney’s guardrails failed more often,” CCDH said in a report, adding that the tool failed in 40 per cent of test cases. By comparison, CCDH said, ChatGPT failed only about 3pc of the time.
CCDH tested the platforms against prompts relating to Biden and Trump as well as French President Emmanuel Macron, German Chancellor Olaf Scholz and European Commission President Ursula von der Leyen.
Midjourney failed in half of all tests relating to Biden and Trump, the report said. Those included an image of Biden being arrested and Trump appearing next to a body double.
Midjourney did not respond to a request for comment.
In March, tech activists reported that Midjourney had blocked all prompts related to Trump and Biden, effectively barring users from creating fake images. CCDH said users could easily circumvent the policy — in some cases by adding a single backslash to a prompt previously blocked by Midjourney.
