DuckDuckGo AI spreads a wild false claim about the U.S. President

The digital world just witnessed one of the wildest AI search failures in history. Anyone who thought AI hallucinations had peaked was proven completely wrong by privacy-centric search engine DuckDuckGo. The platform’s built-in AI assistant, Duck.ai, began telling users that the President of the United States died from a rabies infection.

This incident shows how easily modern AI search features can be tricked by internet trolls, completely destroying user trust. It serves as an alarming reminder that large language models still have absolutely no concept of what is actually true. Instead of a reliable tool, users got a front-row seat to a machine eating its own tail.

A bizarre case of rabies and superpowers

What exactly did the AI claim happened to President Donald Trump? The AI search feature boldly informed users that Trump tragically passed away after succumbing to a severe rabies infection. But the bizarre tale did not stop there.

The AI went on to claim the infection was passed to him by Vice President J.D. Vance, who also supposedly died of rabies shortly before Trump. According to the generated search result, Trump got bitten by Vance on purpose after acting on the advice of Secretary of Health and Human Services Robert F. Kennedy Jr. Supposedly, Trump believed the rabid bite would grant him literal superpowers.

To make things even funnier, the AI claimed Vance originally contracted the virus from a feral, ‘mis-upholstered ottoman.‘ It is the kind of absurd science fiction that is impossible to make up, yet the AI presented it as an absolute fact.

How a satirical subreddit poisoned the well

As anyone might guess, none of this is actually real. It turns out this ridiculous story was a highly successful, coordinated attack by anti-AI activists. The entire narrative originated on a satirical subreddit called r/poisonai.

The members of this subreddit deliberately post fake news to feed hungry AI web scrapers. Their goal is to poison the data pools that AI companies steal to train their models, and in this case, the prank worked like a charm.

What made the situation worse was a recursive feedback loop involving a fake local news site called WKNA News. This automated scraper site copied the Reddit posts, and DuckDuckGo’s AI saw the site’s article as ‘corroboration‘ for the claims. So, the AI essentially cited AI-written slop based on Reddit jokes, creating a closed loop of absolute garbage.

The alarming statistics of AI search failures

While this rabies saga is incredibly funny, it highlights a much broader, systemic problem. AI search features are confidently wrong far more often than we realize, presenting total falsehoods with the most confident language imaginable.

A groundbreaking study by Columbia Journalism Review’s Tow Center for Digital Journalism proves just how bad the problem is. Researchers tested 1,600 queries across eight major generative search tools and found that AI search engines cite incorrect sources 60% of the time. Rather than admitting they do not know something, these models frequently invent highly plausible lies.

The error rates among the industry’s biggest players are shocking. These are massive failure rates for tools that are actively trying to replace traditional search. This is a huge issue because Americans now use AI models as alternatives to Google. People are offloading basic fact-checking to machines that can’t tell the difference between a real news article and a Reddit post.

The corporate spread of vibe citing

This trend of generating highly plausible but entirely fake references is known as ‘vibe citing.‘ Coined by investigators at GPTZero, the term refers to an AI stitching together fragments of real sources, creating the perfect illusion of accuracy while linking to broken pages or irrelevant content.

Vibe citing isn’t just happening to casual searchers; it’s infiltrating professional reports, too. GPTZero recently ran audits on major cybersecurity and business reports from industry giants like Deloitte, Ernst & Young, and KPMG, finding they were stuffed with AI hallucinations. For instance, in EY Canada’s report on cyber threats, the vast majority of the cited URLs were completely broken or fake.

Similarly, in a KPMG report on agentic AI, only 5 out of 45 citations actually pointed to real, accurate sources. The remaining 40 references were completely ‘vibed’—heavily distorted or outright invented by AI research tools. As tech analyst Carmi Levy warned on Ottawa at Work, relying blindly on AI search features carries massive credibility risks.

When prestigious consulting firms publish hallucinated data, they poison the digital pool that other AI models rely on. If multibillion-dollar firms cannot keep AI from hallucinating their briefs, search engines have no hope of keeping their platforms safe from targeted internet trolls.

The ultimate double irony for DuckDuckGo

This brings the story to DuckDuckGo’s ultimate facepalm moment. The privacy-first company had been winning big by marketing itself as the ‘No AI‘ alternative to Google, even launching a brand-new browser extension explicitly pitched to block AI-cluttered slop from search results.

This smart marketing move worked wonders, sparking a 30% uptick in installs of its flagship app as frustrated Google users fled AI Overviews. Yet, at the exact same time, DuckDuckGo kept its own AI search and chat tool active, using third-party models like Anthropic’s Claude 4.5 Haiku and OpenAI’s GPT-5.4 mini that ended up scraping the poison Reddit threads.

By trying to play both sides, running an anti-AI campaign while hosting an AI search tool, DuckDuckGo set itself up for an embarrassing own goal. It is a classic case of corporate FOMO getting the best of a good brand. As one Reddit user wittily observed, large language models are essentially random number generators with their feet on the scale to make the randomness sound pleasing.

The quick reality check

AI search engines are amazing at sounding smart, but they are incredibly bad at figuring out what is actually true. If a small group of Reddit pranksters can convince a major search engine that the U.S. President died of rabies from a feral ottoman, the technology has a major verification issue. Before relying on AI chat for news, remember that a whopping 60% of AI search citations are incorrect, so always check the sources directly.

Disclaimer – This list is solely the author’s opinion based on research and publicly available information. It is not intended to be professional advice.

Like our content? Be sure to follow us