Researchers Show That Hundreds of Bad Samples Can Corrupt Any AI Model
A study found that just 250 poisoned documents were enough to corrupt AI models up to 13 billion parameters in size, showcasing the need for new kinds of defenses.
It turns out poisoning an AI doesn’t take an army of hackers—just a few hundred well-placed documents. A new study found that poisoning an AI model’s training data is far easier than expected—just 250 malicious documents can backdoor models of any size.
The researchers showed that these small-scale attacks worked on systems ranging from 600 million to 13 billion parameters, even when the models were trained on vastly more clean data.
The report, conducted by a consortium of researchers from Anthropic, the UK AI Security Institute, the Alan Turing Institute, OATML, University of Oxford, and ETH Zurich, challenged the long-held assumption that data poisoning depends on controlling a percentage of a model’s training set.
Instead, it found that the key factor is simply the number of poisoned documents added during training.
No specific quote available in the text
Author's summary: AI models can be corrupted with just 250 poisoned documents.
More News
- Brendan Gallagher: No-one is immune to this awful disease
- 12 Fall Dessert Spots In Massachusetts To Add To Your Weekend List
- Universal Sets 2027 Release for Mysterious New Film From 'Everything Everywhere' Directors The Daniels
- NHL Rumors: Lane Hutson and Montreal Canadiens Influenced By Connor McDavid Extension
- Liberty Hall to reprise its role as an opera house for centennial anniversary screening with live piano accompaniment
- 1st Test: Noman Ali’s 10-for seals 93-run win for Pakistan over South Africa
- Dodgers have drunk the Brewers’ milkshake so far
- Ohio judge blocks DeWine’s ban on intoxicating hemp products for 14 days - The Tribune
- Martin won't give up on Stawell dream, but for now the 400m is his focus
- TVA faces nuclear outage crisis - Nuclear Engineering International