Prompt adware has been identified by researchers as a new method for modifying artificial intelligence agent instructions to insert unauthorized advertisements, with further coverage provided by The Hacker News.
Check Point Research disclosed details regarding a jailbreak bundle named Haiou, which targets OpenAI Codex. The bundle modifies configuration files to inject advertisements for a third-party large language model gateway. By utilizing prompt injection techniques such as role-playing and reframing requests into Capture The Flag scenarios, the tool bypasses safety guardrails. Once active, the agent triggers advertisements for a specific website whenever common greetings are used. The bundle, hosted on GitHub, is marketed as a local enhancement package and has been integrated into other open-source tools. Experts note that because these modifications occur through natural language instructions rather than traditional executable payloads, they are difficult for security teams to detect. This development highlights a growing risk in the AI supply chain, as third-party skills and configurations can be manipulated to influence agent behavior. As AI agents are increasingly adopted in enterprise environments, researchers warn that these components must be treated as part of the critical trust chain to prevent potential security compromises.
Source: The Hacker News
