Study finds more prompt injection training data can make AI agents easier to hijack
Researchers from the University of Waterloo, Zhejiang University and KTH found that AI agents retrained on a larger set of prompt injection attacks became easier to hijack in three of four tests. The red-teaming tool that found the most working attacks produced the weakest defense.
Why it matters
Teams securing AI agents should focus on attack variety, not just volume, when testing against prompt injection.
aii.news picks and summarizes stories from trusted news sources with the help of AI. Every story links to the original reporting. How we work.