Low Cost Data Poisoning Attacks Target Open Weight AI Models
A recent demonstration shows that open-weight AI models can be compromised for under $100 using just a handful of malicious training examples.
Editorial Staff
1 min read
Updated about 6 hours ago
In a groundbreaking demonstration, a cybersecurity researcher has shown that open-weight AI models are vulnerable to data poisoning attacks that can be executed for less than $100.
The researcher successfully manipulated a model using as few as ten malicious training examples, highlighting the ease with which these attacks can be carried out.
Remarkably, the entire process can be completed in under an hour, raising significant concerns about the security of open-weight AI systems.