API
< Back to all clusters
[TECHNOLOGY] · United States, China · 8 sources

started · updated

Check Point Research identifies prompt adware in AI jailbreak bundles

Check Point Research has identified a prompt adware mechanism within the Haiou jailbreak bundle, which targets OpenAI Codex. The bundle is designed to weaken model guardrails to facilitate tasks such as hacking and software cracking.

Upon installation, the bundle modifies Codex configuration files to inject advertisements for the author’s website into subsequent agent responses. These advertisements promote an opaque, OpenAI-compatible LLM gateway, which serves as middleware between applications and LLM providers. This appears to be a business model using free jailbreak tools as an acquisition channel for a paid service.

Researchers observed the jailbreak being utilized by a cybercrime actor alongside other tools, including Cobalt Strike, the VulnClaw agent, and the Ladon framework. A misconfigured webserver belonging to a threat actor was found hosting these tools, with logs indicating the compromise of various private computers belonging to Chinese-speaking individuals.

Entities

Check Point Research