The US search engine DuckDuckGo was recently confronted with a massive error in its AI assistant. When asked about the death of the American politician Donald Trump, the system provided the (fictitious) answer that he died of rabies at the beginning of June 2026, as Futurism reports.
According to the response generated by the search engine, the incident was preceded by an alleged rabies infection of politician JD Vance, which is said to have bitten the presidential candidate. This completely made-up story was presented as fact by the artificial intelligence and backed up with seemingly reputable sources, including made-up advice from politician Robert F. Kennedy Jr. He was said to have actually advised people to become infected with rabies. This gives “superpowers”.
The danger of data poisoning
The origin of this misinformation lies in a targeted attack on the training data of the language models, an approach that is known in the professional world as data poisoning. Users of the Reddit platform had organized themselves in a special forum to publish masses of absurd posts about the fictitious death of politicians.
The goal of this group, which has over 45,000 members, is explicitly to mislead the error-prone AI models and cause the automated systems to treat fake news as fact. Due to the sheer volume of coordinated hoaxes, web crawlers mistakenly classify this content as relevant and truthful.
The role of automated news sites
This manipulated data is then picked up by dubious, purely AI-generated news sites that appear to be genuine local media. These platforms process the attempts at manipulation from the Internet forums into seemingly well-founded articles and thus completely obscure the origin of the story.
Since large AI search assistants often uncritically include such supposed news portals in their index, the circle of targeted misinformation closes. The search engines ultimately verified their false statements with articles that were specifically generated by other automated systems based on the original false statements.
The enormous difficulty for the AI models at this point is to accurately detect sarcasm or deliberate deception in human texts. Without reliable parameters for the credibility of a domain, the algorithms often treat a serious news article as completely equivalent to an elaborate forum joke.
The reactions of search engine operators
The DuckDuckGo company responded quickly to the reporting and removed the erroneous answers from its search assistant. In a statement on the Reddit platform, the provider admitted that the system had been deliberately tricked and announced technical improvements to prevent such incidents in the future.
The browser developer Brave Software, whose AI system had similarly fallen for the fake data, was much more defensive. The company defended the technology to Futurism magazine, saying “search engines, with or without AI, are not oracles of truth,” and noted that users still need to use common sense.
The company also pointed out that the result set would automatically adjust as soon as verified leading media reported on this intentional manipulation. This argument shifts the fundamental responsibility for correct information almost completely from the developers to the people in front of the screens.
Weaknesses in quality control
This incident highlights a central problem with modern AI systems that rely on so-called retrieval-augmented generation to pull current information from the Internet and output it in real time. If the algorithms cannot independently distinguish between satirical or intentionally manipulated forum posts and verified news, the general reliability of the entire application will be severely compromised.
Although such models can make it possible to answer complex search queries much more conveniently, a closer look shows them to be extremely vulnerable to coordinated attempts at manipulation. Programmers are therefore faced with the massive task of designing more robust filter mechanisms that can reliably assess not only the textual relevance but also the actual facts of a source.
Top Article
${content}
${custom_anzeige-badge}
${custom_tr-badge}
${section}
${title}
A critical look into practice
The current case mercilessly proves that blind trust in automated search assistants is currently completely inappropriate and that a critical review of the results displayed remains essential. Even though the systems often provide correct summaries in everyday life, the astonishing ease of this manipulation highlights the enormous dark side of current technology.
As long as the underlying mechanisms for evaluating sources are not fundamentally and structurally revised, such outliers in information gathering will probably not remain uncommon. For search engine users, this specifically means that they have to question every answer, while technology providers have to continually prove their full-bodied promises of intelligent search.
