Back to AIBriefs
AnalysisCybersecurityPolicy

AI agents remain vulnerable to prompt injection attacks, study finds

AI agents remain vulnerable to prompt injection attacks, study findsDecrypt

Direct prompt injection attacks succeeded more than 79% of the time against agents powered by GPT-5 and Gemini. Researchers developed StakeBench, a new benchmark to characterize the nuanced victim-dependent risks of such exploits.

·
Jun 12, 7:22 PM
AI agents remain vulnerable to prompt injection attacks, study finds — AIBriefs