OpenAI's latest AI model, GPT-5.2, has been found citing an online encyclopedia called Grokipedia in its responses on sensitive topics, casting doubt on the credibility of the model for professional use.
According to recent tests conducted by The Guardian, when it comes to specific and often contentious subjects such as Iran or the Holocaust, ChatGPT relies heavily on Grokipedia for its answers. This was particularly noticeable when it came to questions related to the Iranian government's alleged ties to MTN-Irancell, a telecommunications company, and Richard Evans, a British historian who testified in a libel trial involving Holocaust denier David Irving.
However, The Guardian discovered that ChatGPT does not cite Grokipedia when responding to more general or non-sensitive prompts. Instead, it relies on its own internal knowledge base and internet search results for information on these topics.
Grokipedia itself has faced controversy due to the inclusion of citations from neo-Nazi forums in some of its articles. A study conducted by US researchers has also raised concerns about the reliability of the encyclopedia's sources, labeling them as "questionable" and "problematic".
In response to The Guardian's findings, OpenAI claimed that its GPT-5.2 model searches a wide range of publicly available sources and viewpoints but applies safety filters to minimize the risk of surfacing links associated with sensitive or harmful content.
The discovery raises questions about the ability of AI models like GPT-5.2 to provide trustworthy information on complex and contentious topics, particularly when they rely on external sources such as Grokipedia.
According to recent tests conducted by The Guardian, when it comes to specific and often contentious subjects such as Iran or the Holocaust, ChatGPT relies heavily on Grokipedia for its answers. This was particularly noticeable when it came to questions related to the Iranian government's alleged ties to MTN-Irancell, a telecommunications company, and Richard Evans, a British historian who testified in a libel trial involving Holocaust denier David Irving.
However, The Guardian discovered that ChatGPT does not cite Grokipedia when responding to more general or non-sensitive prompts. Instead, it relies on its own internal knowledge base and internet search results for information on these topics.
Grokipedia itself has faced controversy due to the inclusion of citations from neo-Nazi forums in some of its articles. A study conducted by US researchers has also raised concerns about the reliability of the encyclopedia's sources, labeling them as "questionable" and "problematic".
In response to The Guardian's findings, OpenAI claimed that its GPT-5.2 model searches a wide range of publicly available sources and viewpoints but applies safety filters to minimize the risk of surfacing links associated with sensitive or harmful content.
The discovery raises questions about the ability of AI models like GPT-5.2 to provide trustworthy information on complex and contentious topics, particularly when they rely on external sources such as Grokipedia.
this is so worrying its like how can we trust these AI models to give us accurate info especially on sensitive topics like that 
what if it gets hacked or biased in some way? 
i mean i get it openai trying to minimize the risk of surfacing bad content but it seems they still got caught out 
grokipedia itself is a red flag, who uses online forums for citations lol 
i guess this just goes to show how much we need to work on our critical thinking skills when using these AI tools 
. And yeah, maybe Grokipedia isn't the most reliable source, but that just means we're pushing AI to be more diverse and nuanced in its knowledge base, right? 
It's like they say: "the devil is in the details"... or in this case, the questionable sources on Grokipedia 
. But seriously, it's a great reminder that AI models are only as good as their training data, so let's hope OpenAI takes steps to improve its safety filters and ensure we're getting accurate info from our tech 
.
. I mean, who wants to read info from neo-Nazi forums? Not me, that's for sure 
.
. We can't just assume that because they've got some fancy safety filters, everything will be okay 
.
. I've been using GPT-5.2 for my personal projects and was actually pretty impressed with how well it handles general questions. But now that I think about it, it makes total sense that it would rely on Grokipedia for sensitive topics - those guys have some sketchy sources, you know? 
.
Still, I think this is a great opportunity to teach people about fact-checking and critical thinking... we need those skills more than ever! 
but omg the fact that it only uses grokipedia 4 sensitive topics is wild 
cant say im surprised tho cuz we all no how sketchy grokipedia is 
anyway, this just shows us how much we dont no about AI and its capabilities... maybe its time 4 more research on this topic? 
it's like OpenAI is trying to spin this as some kind of solution when really they're just exposing their own AI model for being flawed 
I mean, if GPT-5.2 is gonna cite some sketchy online encyclopedia like Grokipedia for super sensitive topics, what's next? Are we gonna be getting fake news from our AI pals too?! 
And don't even get me started on the fact that Grokipedia has its own set of problems with questionable sources... it's like a never-ending nightmare! 