Check Point Research reported on 3 November 2026 that Haiou, a jailbreak bundle targeting OpenAI Codex, persistently modifies agent instructions to advertise its author's large language model (LLM) gateway. Its analysis covers version 4.5.0 and classifies the behavior as prompt adware: natural-language changes to trusted instructions that influence future responses for commercial or behavioral objectives.[1]

According to the report, the bundle installs six security-related Skills and modular prompts, including instructions intended to suppress refusals and reframe offensive-security requests as educational exercises. An identity module directs the agent to answer specified greetings with a gateway advertisement. Another module designates the promoted service as protected infrastructure and forbids offensive testing against it, even when a user claims authorization.[1]

The launchers reportedly check Ed25519 signatures and SHA-256 hashes, then decrypt a packaged payload using an embedded AES-256-GCM key. Those checks concern the package; they do not establish that its instructions match a user's expectations. Check Point says encrypted packaging obscures the prompts from casual repository inspection. It also reports an uninstaller that restores original configurations and removes the advertisement, but identifies no fixed release or recommended upgrade version.[1]

Check Point found the bundle on a cybercrime actor's exposed server. It explicitly says this does not establish that the author created it for cybercrime, and reports no indication that the advertised gateway itself is malicious.[1]

[1]: Check Point Research, “Free the Model, Sell the Gateway: Prompt Adware hidden in Jailbreak Bundles,” 3 November 2026. https://research.checkpoint.com/2026/free-the-model-sell-the-gateway-prompt-adware-hidden-in-jailbreak-bundles/

Sources

  1. Free the Model, Sell the Gateway: Prompt Adware hidden in Jailbreak Bundlesresearch.checkpoint.com · 3 Nov 2026

Check or correct this page

Verification record

The digest below shows whether the text you are reading matches the published source. It does not show that the reporting is correct. The build computes it from this page's Markdown source and lists it in the site manifest.

43e8c7987a87542117493e989c000ec03434b49ca1dda383ee565b5d31c80162