Researchers find prompt injection attacks defeat safeguards across GPT-5 and Gemini agents
According to CSO Online, research from Nanyang Technological University (StakeBench) found that no single attack scenario was consistently blocked across leading AI web agents powered by GPT-5 and Gemini, indicating current systems lack dependable defenses against prompt injection. The study tested multiple AI agent implementations and found systematic gaps in how they handle adversarial inputs designed to manipulate agent behavior.
Topics
Sources
- Press Read article
Go deeper
This intelligence is sourced automatically from public sources across the web and synthesised by the Prefactor AI pipeline. Stories are reviewed before publication.