Posted on 10/10/2025 3:25:23 AM PDT by C19fan
Poisoning AI models might be way easier than previously thought if an Anthropic study is anything to go on.
Researchers at the US AI firm, working with the UK AI Security Institute, Alan Turing Institute, and other academic institutions, said today that it takes only 250 specially crafted documents to force a generative AI model to spit out gibberish when presented with a certain trigger phrase.
For those unfamiliar with AI poisoning, it's an attack that relies on introducing malicious information into AI training datasets that convinces them to return, say, faulty code snippets or exfiltrate sensitive data.
The common assumption about poisoning attacks, Anthropic noted, was that an attacker had to control a certain percentage of model training data in order to make a poisoning attack successful, but their trials show that's not the case in the slightest - at least for one particular kind of attack.
(Excerpt) Read more at theregister.com ...
Same thing happens to democrats when you interject the truth or better yet just mention President Trump.
Garbage in, garbage out. I’ve stated numerous times AI’s learning ability is similar to kids in school. It only learns what it’s allowed to access. The AI is poisoned just like kids minds in school.
Now THAT was funny! Bravo!
Same thing happens to democrats when you interject the truth or better yet just mention President Trump.
/
πππ
SO true.
Total reliance on anything AI is foolish. It is much like WikiPedia - everything should be read with caution. If you see an AI response to a question, first ask yourself “Is that reasonable?” Example: many in the leftist media say and have said for years that there are 11-12MM illegals in the USA. Anyone with a triple-digit IQ knows that number is not reasonable.
For Tom
Gibberish? Like the headline?
BINGO!!
This is the key bit of information. GIGO has an effect on ANY information gathering/dispensing system, and AI is especially so because the “Intelligence” involved is so . . . artificial.
Who programmed it, and who feeds it what data.
Define “illegal”. If Biden put 3 million in “parole” status their status is no longer “illegal”. If an H1b worker is expecting to marry a citizen and get a new visa based on that and is in “transition” he is not an “illegal”. If an “illegal” successfully achieves another status than he is no longer “illegal”.
The definitions are all over the place.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.