Haiou Codex jailbreak toolkit found injecting adverts into AI responses
- Severity
- Medium
- Sources
- 1
What happened
Check Point's November 3 report describes Haiou 4.5.0, a Chinese-language OpenAI Codex enhancement distributed via GitHub (account haiou-666, repository haiou-codex), which the researchers classify as prompt adware. The payload (payload.pkg) is encrypted with AES-256-GCM and verified via Ed25519 and SHA-256, and it includes jailbreak modules (role-play, refusal suppression) plus skills such as seagull-pentest and seagull-reverse. Defenders should review 00-identity.mdgreater, CLAUDE.md, installed skills and gateway settings (zz.ai) before trusting enhancement bundles; no CVE is involved. The package was found alongside Ladon and VulnClaw on an exposed Cobalt Strike server, but no link to cybercrime by the author was established.
For security analysts · Written by AI from 1 source · Wrong?
What to do
- Review prompt instruction files, installed skills and gateway settings before trusting any Codex enhancement bundle.
- If Haiou is installed, use its included uninstaller, which restores original configurations and removes the advertisement.
- Treat readable instructions in third-party AI tool packages as configuration that needs review, not harmless text.
Written by AI from the sources, checked against them
Full story timeline · 1 update
Who reported · 1 outlet
Who and what is in it
Malware
Companies