Observed incident
AIR reports live credential exposure from incorrect Render skill guidance
Reported
What happened
AIR Security found incorrect networking guidance in an official deployment skill. Codex reproduced the domain in generated apps, and AIR says real applications sent credentials to its passive sinkhole.
Parties and reporting
AI tools and systems: OpenAI Codex, render-deploy skill, Render
Outcome
AIR reports an August 14 Render correction, August 26 OpenAI plugin removal and transfer of the domain to Render. This does not establish remediation of previously generated applications or all skill copies.
Operating lesson
Validate domains in agent instructions and generated configurations; update affected skills, correct deployed endpoints and rotate potentially exposed credentials.
Evidence and limits
Researcher-reported accidental credential exposure, not a verified criminal campaign. AIR provides no victim count. October 14 dates its security report; August fixes predate it, and October 16 is later reporting.
Sources
- AIR Security OpenAI Sesame disclosure
Primary source ·
- Render correction to internal-address guidance
Primary source ·