Threat
Shadowy AI Relay Disables OpenAI Codex, Claude Code Sandboxes
First reported darkreading.com
Page published
Earliest dated coverage: 29 Oct 2026 · First observed: 29 Oct 2026 · Latest dated coverage: 29 Oct 2026
Coverage timeline
Single-source incident — one report is available.
Why it matters
The relay weakens containment of coding-agent actions while placing enterprise AI requests under an untrusted intermediary’s control.
An unauthorized AI relay reportedly installs genuine OpenAI Codex and Anthropic Claude Code command-line tools on enterprise systems, disables their sandboxes, and redirects requests through its own infrastructure. Symantec researchers report that the service has operated since early 2026, providing access in regions where the vendors do not officially offer their products.