A tiny snippet of user-generated text as short as 13 words long is often enough to manipulate the AI agents that power tools like ChatGPT and Google’s AI search, new research shows. The study suggests that it is trivially easy for brands to inject promotional content on sites like Reddit, Quora, and Wikipedia with the end goal of poisoning or manipulating the output of AI tools.

The preprint research, done by Hal Triedman, Tingwei Zhang, and Vitaly Shmatikov of Cornell University, is called “Deep-research agents can be poisoned via user-generated content” and provides a mechanism and research basis for a problem that has been noticed by Reddit moderators and Wikipedia editors, namely that their websites are getting flooded with promotional content from brands trying to do AEO, or AI-engine optimization. 404 Media has repeatedly reported on this booming industry, in which brands try to promote their product by seeding the websites that AI tools most often cite and scrape from with inauthentic and spammy content.

The Cornell research finds that deep research agents, which are the real-time scrapers that tools like Google AI search and ChatGPT use to retrieve web content with citations in response to user queries, cite user-generated content from sites like Reddit or Wikipedia in roughly half of all queries, and that nearly a quarter of all citations come from user-generated websites. The paper suggests that what we have been seeing is basically Redditor suggests you put glue on your pizza as a service, or an end-to-end attack against the systems that increasingly dominate the ways that people access information online. The researchers found that “a single poisoned Reddit comment can influence generated outputs for an entire cluster of related [AI] queries,” the paper said.

“We show that a tiny snippet—just 13 words—of retrieved text on a UGC website like Reddit, Wikipedia, Quora, Facebook, etc. can change AI agents to output spam / scam content pretty consistently,” Triedman told 404 Media.

1 comment

sorted by: hot top controversial new old
[–] 1 point 4 months ago

Yeah, we know. Redditors deliberately added the glue in the cheese recipe to the prompt for keeping it from sliding it off pizza.

  • source
  • 15 comments in c/fuck_ai@lemmy.world view that post
    [–] 26 points 4 months ago (8 children)

    This is happening on Lemmy. There's a bot on technology that "finds and expands abbreviations" but somehow always finds CloudFlare even if it isn't mentioned or even related to networking.

    So "Cloudflare" effectively SEO'd any popular Lemmy thread.

  • source
  • hideshow 8 child comments
  • [+] 25 points 4 months ago (7 children)
  • [–] 19 points 4 months ago (4 children)

    I feel like it says something about the state of the internet that I wasn't sure if this was ironic or legitimately a bot until I checked your profile

  • source
  • parent
  • hideshow 4 child comments
  • [–] 13 points 4 months ago

    Remember this is also why they aggressively censor anti-genocide rhetoric. They don't want their AI influenced.

  • source
  • [–] 8 points 4 months ago (1 child)

    Fuck Reddit and Fuck Spez.

  • source
  • hideshow 1 child comment
  • [–] 12 points 4 months ago

    Aside from the clownshow that is Gemini, I've found my own ye olde reddit threads fill the AI search result with useless information I wrote years ago lol.

    If ti weren't for the fact that they've ruined everything with AI, it would still be funny seeing it present meme logic as the authentic truth.

  • source
  • [–] 4 points 4 months ago

    We've yet to even talk about data quality in training these things.

  • source
  • [–] 2 points 4 months ago
    9 comments in c/technology@lemmy.world view that post
    [–] 17 points 4 months ago* (2 children)

    Is this really an attack? They use a well known domain to claim banancoin is a fast growing crypto and the result talks about established coins and mentions their poisoned result at the end. But an article published on any well known domain would do the same thing to a regular search engine no?

  • source
  • hideshow 2 child comments
  • [–] 11 points 4 months ago (1 child)

    Regular search engines have feedback mechanisms that limit how effective that is. The click through and bounce rate are used to adjust rankings and as more and more people look at the fake info and then ignore it it will naturally fall out of the top results and get buried. LLMs though don't have that feedback, once something is ingested and baked into the model it's there forever. The fake info doesn't need to look believable enough to fool a human, just self consistent enough to fool an LLM with its tiny context window.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 12 points 4 months ago (1 child)
  • [–] 8 points 4 months ago

    LLMs seem to love reddit a lot. Often times I find they confidently state information with the sourcing being a single reddit comment on a years old post

  • source
  • [–] 6 points 4 months ago

    I did my part by lying about shit and insulting users who debate me for years.

    If your AI says "Fuck you!" when you question it's answer... You are welcome!

  • source
  • [–] 5 points 4 months ago

    Doctor Who: Don't you think (she, the PM) Trump looks tired?

  • source
  • [–] 3 points 4 months ago

    We have sophisticated systems that detect and prevent inauthentic behavior, coordinated manipulation, and astroturfing, and we recently announced that any fishy automated accounts will be asked to verify their humanity.

    Uh huh sure. Like I didn't just watch my favorite gaming subreddit get locked after 2 weeks of constant copy paste spam posts about IPTV that even the simplest automod line could have fixed. I reported every single one of the and yet the accounts could keep posting.

    They weren't AI, but rather paid spammers from spam networks in India or Africa, so they can bypass biometric requirements.

  • source
  • 8 comments in c/reddit@lemmy.world view that post
    [+] 13 points 4 months ago* (4 children)
  • [–] 9 points 4 months ago (2 children)

    So dumb anybody will question it, not parallel enough to the truth it becomes conspiracy.

    Just as general warning, I'll point out that there are a fair number of people who believe that the Earth is flat.

  • source
  • parent
  • hideshow 2 child comments
  • load more comments (1 reply)
  • [–] 6 points 4 months ago (2 children)

    Wow just 1 short of the average reddit comment these days

  • source
  • hideshow 2 child comments
  • 1 comment in c/technology@piefed.social view that post
    [–] 5 points 4 months ago

    What's funny is that Reddit has been heavily infested by bots and trolls even before widespread LLM usage for chatbots. And it became even more evident after certain subs became dead overnight after Reddit got banned in Russia.

    And now it has come to just a cycle of AI data regurgitation and eating it back up. 🌚

  • source