Check Point Identifies Haiou Prompt Adware Injecting Ads into Codex Global Instructions via GitHub Bundle
Haiou represents the first documented case of monetized prompt adware delivered through a public jailbreak toolkit. The attack surface shifts from code execution to persistent instruction poisoning in AI agents. Organizations integrating third-party Codex or agent configs must treat prompt files as untrusted supply-chain components.
Official Check Point reporting stops at the ad-delivery mechanism. Independent analysis of the repo history reveals additional reverse-engineering and cracking workflows bundled alongside the ads, expanding the risk surface beyond nuisance. Enterprise agents pulling unvetted Codex configs inherit these instructions as trusted system prompts. Next indicators will be increased forks of similar 'personality' packages and attempts to port the technique to other local agent frameworks.
Sources (2)
- [1]Check Point Research Haiou Report(https://research.checkpoint.com/haiou-prompt-adware)
- [2]GitHub haiou-codex Repository(https://github.com/haiou-codex/haiou-codex)