逆向发现 Muse 后台代理疑似调用 OpenAI

Jurly @jurlycat原推发布于 2026.09.26

在 X 上打开原推

Meta built Muse around its internal Avocado models. Then a reverse engineer found a background agent calling: azure/muse-special The traces looked suspiciously OpenAI-shaped: • call_ tool IDs • gpt_responses_v1 objects • Encrypted reasoning payloads • Runtime clients for Azure GPT, Claude and Kimi Hacker News is now debating whether this is an actual OpenAI model or just a Meta model using an OpenAI-compatible API. There’s no definitive proof yet. But the debate reveals a bigger problem: Once an agent hides inference behind a private runtime, users have almost no way to know which model is actually doing the work.