A jailbreak is a technique for bypassing the security restrictions of an AI model using a specially crafted prompt, while prompt injection involves embedding hidden instructions into external content (such as a website or document) with the aim of taking control of the behavior of the model processing that content. Both techniques pose a serious security risk, particularly when deploying AI agents that autonomously process external, untrusted data.
Jailbreak / prompt injection
The AI Act for Board Members: What Every C-Level Executive Needs to Know to Keep the Company from Paying Millions
6 min read
AI Governance in the Workplace: How to Set Rules for AI Use Before Regulators Do It for You
5 min read