Shadowy AI Relay Disables OpenAI Codex, Claude Code Sandboxes
2026-10-29
Dark Reading
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
A grey market relay service installs genuine OpenAI Codex and Anthropic Claude Code tools at 12 organizations, mainly in mainland China, disabling sandboxing and rerouting data through third-party servers. This exposes sensitive enterprise information and disables security protections, creating significant risks and realized harm to affected companies.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves AI systems (OpenAI Codex and Anthropic Claude Code) whose use has been altered by a relay service that disables security features, creating a credible risk of harm to organizations using these tools. Although no direct harm or malicious activity has been observed, the disabling of execution sandboxes and replacement of credentials significantly increase the risk of cyberattacks or data breaches. This fits the definition of an AI Hazard, as the development and use of the AI systems in this unauthorized and insecure manner could plausibly lead to an AI Incident involving harm to property or communities. The event does not qualify as an AI Incident because no realized harm has occurred yet, nor is it Complementary Information or Beneficial Use.[AI generated]