AI Stack Alert Medium TD-2026-0007 For CISO, SOC / IR, AppSec, AI Eng AI-assisted
Prompt adware "Haiou" hidden in OpenAI Codex jailbreak bundle injects ads
Check Point Research reported a jailbreak bundle named Haiou that targets OpenAI Codex. The bundle modifies Codex configuration and global agent instruction files so that common greetings trigger advertisements for a third-party OpenAI-compatible LLM gateway. The researchers describe this behavior as "prompt adware."
Why it matters
Organizations that install third-party Codex skills or configuration packages can have agent instructions changed without any executable being run. This can add persistent promotional content and creates a supply chain path for other unwanted agent behavior.
Do this
- Inventory Codex installations and review global agent instruction files and Codex configuration for unexpected advertising text or unknown gateway URLs, such as zz.aiapi2025[.]top.
- Block or alert on the domain zz.aiapi2025[.]top in DNS and web proxy logs, and check endpoints for connections to it.
- Treat third-party Codex skills, configuration packages, and GUI configuration tools as untrusted supply chain components. Require review before installing them on managed devices.
- Restrict agent use of unapproved LLM gateways and require developers to use approved provider endpoints.
Signals
- NoCISA KEV
- 26/100Priority
- 70%Confidence
Evidence
“designed to modify an artificial intelligence (AI) agent's trusted instructions to insert advertisements for a third-party AI gateway”
“the jailbreak modifies OpenAI Codex's configuration files to inject advertisements for the jailbreak author's website into subsequent agent responses.”
“the findings demonstrate how a bad actor could introduce undesirable behavior through third-party skills or configurations, leading to supply chain attacks.”
Sources
Affected / related
OpenAI Codex, haiou-codex, Codex-X, Check Point Research, prompt adware
Published 2026-11-04 16:06 UTC · updated 2026-11-08 04:31 UTC · AI-assisted summary, checked against quoted sources · How we score