Check Point Research flags prompt adware risk in AI coding agents

Check Point Research has identified a technique it calls prompt adware, in which trusted instructions given to an AI agent are persistently altered to inject advertising for a third-party AI gateway. The researchers traced it to Haiou, a jailbreak bundle targeting OpenAI Codex.
Because the manipulation relies on natural-language prompts and configuration files rather than executable code, it can quietly shape an agent’s future behaviour. Check Point said the advertisement itself is simple, but the method raises questions about how much visibility organisations have into the third-party prompts, Skills and configurations influencing the AI systems they trust.