- Chatgpt cannot say if a site has been hacked, expired or reused for Casino spam
- The answers generated by Ai-Ai may seem reliable, even when they completely cite the diverted and false sources
- Expired areas of charity are reborn like gaming sites and always pass as sources of confidence in AI
Chatgpt quickly becomes an essential source for people looking for recommendations, online services to local businesses, but new evidence suggest that its suggestions generated by AI cannot always be based on reliable sources.
In fact, some are from websites that have been hacked, whose domains have expired and have been reused, often to promote online casinos and game platforms.
In recent months, James Brockbank, managing director and founder of Digitaloft, has documented how the problem has become widespread, by discovering examples of Chatgpt citing the content of the sites that have been clearly manipulated.
Operate the gaps in the validation of the source of AI
In a case, a functional legal practice website, managed by lawyer Veronica T. Barton, has buried pages recommending buried British casinos.
“Their site was hacked and this page added,” noted Brockbank after examining the evidence.
In another case, a site formerly affiliated with a coalition of young people from the United Nations had been transformed into a platform pushing “Casinos not on Gamstop”.
Although the listle he hosts contains only one external link, he led to another reused area.
The model continued with expired areas, including the one belonging to an artistic charitable organization now disappeared previously linked by the BBC, CNN and Bloomberg.
This area, which now pushes the content of the game, was cited by Chatgpt in response to a question on the casinos without deposit.
These tactics exploit the weaknesses of how Chatgpt selects and cites sources, as unlike traditional search engines, the model lack of mechanisms to verify the legitimacy of the property of a site or editorial intention.
Consequently, the content injected on compromise websites can surface in its responses without any obvious red indicator to the user.
Chatgpt seems to promote the recent content and always attributes the authority based on the reputation of the inherited domain, even when the content of the domain has no continuity with its past – which opens the door to the bad actors to manipulate visibility by means which have little to do with credibility.
The main thing is that users turning to Chatgpt for recommendations should not assume that each response is supported by a credible source.
A rapid verification of the authority of the cited site, its history, its property and its relevance can greatly contribute to avoiding deceptive or harmful suggestions.
