Coding harnesses hand shell access to inference providers — 4-step gateway mitigation
shensi · x · 2026-11-12
@abiryanii flagged an underappreciated attack surface: plugging an inference provider into your coding harness effectively gives that provider shell access to your machine — an untapped attack vector, for now.
@yashgogri1 proposed concrete mitigations:
- Run an MCP + LLM Gateway you fully control, governing which tools and models each employee can access;
- Deploy guardrails at both the gateway and harness levels;
- Enforce the gateway as the only allowed path via admin policy;
- Continuously monitor gateway logs and alerts.
Running coding agents in sandboxes is the final step; the first four already give you strong control.
Related event: Coding Agents with Third-Party Inference Risk Handing Over Shell Access(2 posts)→
More from coding & agent
- Dev's favorite AI use case: Claude monitors competitors and emails alerts — doooyle · 2026-11-12
- Claude watches competitors around the clock and emails me what matters — GlenBradley · 2026-11-12
- Developers say Gemini 4 Carbon is notably better at favicon design and micro-interactions — AI_Andrew · 2026-11-12
- v0 now supports ChatGPT Business subscriptions alongside Plus and Pro — OpenAIDevs · 2026-11-12
- Nvidia's OpenShell and Google's Agent Substrate join CNCF — pjausovec · 2026-11-12
- Free ACM TechTalk on Harness Engineering: From Hopeful Instructions to Enforced Control — marlene_zw · 2026-11-12