An LLM gateway isn't a control plane if agents can bypass it entirely
Arc_bong · reddit · 2026-09-09
The author argues that LLM gateways (LiteLLM, Portkey, OpenRouter) only govern traffic that passes through them—agents with credentials and network access can call model APIs, MCP endpoints, or external services directly, at which point the gateway still routes traffic but no longer governs the agent.
Key points:
- Traffic control and path control are different problems; a gateway should be one component of agent governance, not the governance boundary itself.
- A better architecture adds an Agent Gateway layer above the LLM gateway, paired with network egress enforcement, identity, and shadow discovery to detect and close bypass paths (citing Lyzr Open Controller as an example).
- The issue will grow as agent estates spread across Kubernetes, cloud agent runtimes, and MCP servers.
The author closes by asking practitioners what actually prevents bypass in production, beyond what architecture diagrams claim.
Related event: Agents Can Bypass LLM Gateways, Undermining Control-Plane Claims(2 posts)→
More from coding & agent
- Is anyone actually using Meta's Ads MCP for ad automation? — bir_ch · 2026-09-10
- Codex glitch: OpenAI to compensate affected users with extra banked resets — Angaisb_ · 2026-09-10
- Agent swarm probe expands: abused API key, new spam sites, activity until Sept 2 — xeophon · 2026-09-10
- Dev burns 2 days on basic email automation with Google Astra, blocked by CAPTCHAs and logins — AjdDavison · 2026-09-10
- Dev now clears a week of Codex and Claude output in about a day: 'honestly a little scary' — jacob_posel · 2026-09-10
- Filmmaker uses Computer Use to drive Blender and editing, calling taste the new interface — taherdhanera · 2026-09-10