I've been seeing the back-side of this as a mod at the Julia Discourse. Pagerank/SEO spam is ridiculously obvious and trivial to detect/block — links are all that matter. Many GEO spam posts are similarly obvious: someone posting about a cryptocurrency customer support hotline on a programming language forum is definitely spam. But recently spammers/scammers been using AI to _tailor_ posts to look much more authentic, posting "How to use Julia to analyze market statistics" referencing (without links!) a particular crypto exchange and then (sometimes) going back later to edit in phone numbers or the like.
What makes this even more painful is seeing all the good faith answers that such GEO posts spur from the community. It's far more abusive than SEO spam.
That's happening across the board when it comes to models that feed off online information. Create a website with a false claim, have AI slurp it up and spit it back out when a user asks a question.
Oh no, my uncritical ingestion of data, based on the assumption that all data is of equal quality, and that more is better, has failed. How could this happen!
You've always needed to be critical of your sources, your teacher tried to explain that when you ripped of that Wikipedia article or clicked the first link in a Google search. How the fuck did the AI companies think they could avoid reading and rating the content they've been hovering up?
Because their leaderships have defunded / fired all ethics researchers looking into their ongoing crimes and failures in favor of a bs "AI safety" narrative based around a creepypasta about an AI monster time traveling backwards to torture their staff for not maximizing shareholder value.
That's very troubling. The scariest part is how easy it is, one Instagram post and you may change the customer phone number of Airbnb, or who is leading the election poles.
There's an age-old SEO / spamming thing happening here, but it's made worse by the LLMs just being unbelievably credulous. They'll wrap anything up in a veneer of authenticity, and Google's search AI box only adds to that.
I run into this all the time when I'm doing product work. I'll dump a call transcript from a feedback call into Claude, and it'll believe every word. "The user said they'd use this feature, you should build it!" No they won't! The whole point of doing this analysis is trying to separate the genuine information from the conversational niceties, and god the LLMs are terrible at that.
It's happening at the startup sphere too. There are companies now creating fake company pages to recommend and advertise products so that the LLMs can consume them.
It’s not just hackers - since legitimate sites block AI crawlers, the AI are just grabbing whatever will let them read anything - killing legit sites and feeding slop farms and feeding misinformation! The future is so bright!
Did takedowns for a brand's fake support numbers and the hard part was never finding them, it was that killing one PDF just moved it to a new Medium post by morning.
ChatGPT, Gemini, and Google AI Overview are being poisoned by a massive AI disinformation attack. When users look up everyday info of hundreds of major companies, AI is delivering phishing traps disguised as trusted answers.
Attackers are flooding the web with carefully optimized posts, PDFs, reviews, and fake support pages, to trick AI into presenting fraudulent phone numbers, email addresses, and login pages.
The targets included Delta, Lufthansa, Qatar Airways, Chase, Bank of America, Airbnb, TripAdvisor, and hundreds more.
What makes this even more painful is seeing all the good faith answers that such GEO posts spur from the community. It's far more abusive than SEO spam.
Happy elections.
You've always needed to be critical of your sources, your teacher tried to explain that when you ripped of that Wikipedia article or clicked the first link in a Google search. How the fuck did the AI companies think they could avoid reading and rating the content they've been hovering up?
Back at benevolence or malevolence.
I run into this all the time when I'm doing product work. I'll dump a call transcript from a feedback call into Claude, and it'll believe every word. "The user said they'd use this feature, you should build it!" No they won't! The whole point of doing this analysis is trying to separate the genuine information from the conversational niceties, and god the LLMs are terrible at that.
Attackers are flooding the web with carefully optimized posts, PDFs, reviews, and fake support pages, to trick AI into presenting fraudulent phone numbers, email addresses, and login pages.
The targets included Delta, Lufthansa, Qatar Airways, Chase, Bank of America, Airbnb, TripAdvisor, and hundreds more.