- Chatgpt cannot tell if a site was hacked, expired or recycled to casino spam
- AI-Generated answers may seem reliable even when quoting completely hijacked and false sources
- Expired charity domains are reborn as gaming places and still pass as reliable AI sources
Chatgpt quickly becomes a go-to source for people seeking recommendations from online services to local businesses, but new evidence suggests that the AI-generated suggestions may not always be based on reliable sources.
In fact, some are drawn from sites that have either been hacked or whose domains have expired and have been recycled, often to promote online casinos and game platforms.
Over the past several months, James Brockbank, CEO and founder of Digitaloft, has documented how widespread the problem has become, and reveals examples of chatgpt citing content from websites that have clearly been manipulated.
Utilizing holes in AI -kild validation
In one case, a functioning legal practice had website run by lawyer Veronica T. Barton, pages recommended British casinos buried within it.
“Their site has been hacked and this page added,” Brockbank noted after reviewing the evidence.
In another case, a place that was once associated with a UN youth coalition had been transformed into a platform that pushed “casinos not at gamstop.”
Although the list it hosted only contained an external link, it led to another recycled domain.
The pattern continued with expired domains, including one that had belonged to a now closed art organization previously associated with BBC, CNN and Bloomberg.
This domain, which is now pushing game content, was cited by chatgpt in response to a query about no-depoccasinos.
These tactics utilize weaknesses in how Chatgpt chooses and quotes sources that, unlike traditional search engines, lack the model mechanisms to verify the legitimacy of a site’s ownership or editorial intention.
As a result, content that is injected on compromised sites can surface in its answers without any obvious red flags to the user.
Chatgpt seems to favor recent content and attributes still authority based on older domains reputation, even when the content of the domain has no continuity with its past – which opens the door to bad actors to manipulate visibility through funds that have suffered with credibility.
The main points are that users who approach chatgpt for recommendations should not assume that any answer is supported by a credible source.
A quick control of the quoted site’s authority, its history, ownership and relevance can go a long way in avoiding misleading or harmful suggestions.
