Having issues ... error 403

[rakxdev]

using in deepseek harness

[2026-08-29T02:10:23.374Z] [WARN] Upstream 403 body: {"code":410004,"message":"账号已被封禁"}
[2026-08-29T02:10:23.378Z] [ERROR] Upstream error 403: 账号已被封禁
[2026-08-29T02:13:27.633Z] [INFO] chat model=zai_glm-5.3-flash stream=true
[2026-08-29T02:13:27.640Z] [INFO] Token loaded (expires cache in 5 min)
[2026-08-29T02:14:53.530Z] [WARN] Upstream 403 body: {"code":410004,"message":"账号已被封禁"}
[2026-08-29T02:14:53.530Z] [ERROR] Upstream error 403: 账号已被封禁
[2026-08-29T02:14:55.578Z] [INFO] chat model=zai_glm-5.3-flash stream=true

[eequaled]

this ones on AutoClaws side, the proxy sends the error messages that comes from the upstream directly

403 code 410004 账号已被封禁 translates to "account has been banned"

comes from them or ur account

does autoclaw work and the proxy doesnt? did u try sending a message in the app?

What version of the proxy are you running? You filed this before a fix for the authentication of the requests: the proxy was sending a stale client version (1.17.5) to the upstream while AutoClaw has been on 1.17.8 and the backend gates requests on client identity (already root caused the system prompt watermark requirement). A stale version can trigger these "banned" errors even when not true.

update the proxy and tell me if its fixed, its on npm now:

Copy
npm i -g glmproxy
2.6.0 has the client version fix and maps this exact error to a proper one (account_banned) instead of a generic 403. if you still get 410004 after updating, its real, from their side

i wouldve liked to get more details of the situation. what works and doesnt to make sure whats actually going wrong annd if its proxy side or their side. tell me when it works to close the issue

[rakxdev]

this ones on AutoClaws side, the proxy sends the error messages that comes from the upstream directly

403 code 410004 账号已被封禁 translates to "account has been banned"

comes from them or ur account

does autoclaw work and the proxy doesnt? did u try sending a message in the app?

What version of the proxy are you running? You filed this before a fix for the authentication of the requests: the proxy was sending a stale client version (1.17.5) to the upstream while AutoClaw has been on 1.17.8 and the backend gates requests on client identity (already root caused the system prompt watermark requirement). A stale version can trigger these "banned" errors even when not true.

update the proxy and tell me if its fixed, its on npm now:

Copy npm i -g glmproxy 2.6.0 has the client version fix and maps this exact error to a proper one (account_banned) instead of a generic 403. if you still get 410004 after updating, its real, from their side

i wouldve liked to get more details of the situation. what works and doesnt to make sure whats actually going wrong annd if its proxy side or their side. tell me when it works to close the issue

Thank you for the detailed response and for explaining what might have happened.

To give you some more context about my situation:

After I got this error, AutoClaw itself also stopped working. I tried using it directly, but every response was showing a Chinese message saying that the account had been suspended/banned.

At that time I was using glmproxy v2.0.0. Since the account was already suspended, I couldn't really test whether the proxy was the only thing causing the problem because after that nothing worked anymore. Every model was giving the same error.

The 403 / 410004 error is what I saw through the proxy, but the account suspension was also visible directly in AutoClaw, so it wasn't only a proxy error.

One more detail that might be useful: I had been using the proxy for around 2 hours continuously, and during that time I had consumed more than 8,000 AutoClaw points/credits before this happened.

I haven't tested v2.6.0 yet. I'll update the proxy and test it properly. If I get the same error again, I'll report back with the details so we can figure out whether it's actually a proxy-side issue or something coming from the upstream.

I'm not sure what caused the suspension, but right now it seems like it may have happened on the server/upstream side, possibly because of some account or client checks. I can't say that for certain though.

Thanks again for looking into this and for fixing the client version issue. I'll test the new version and let you know what happens.