ChatGPT icon indicating copy to clipboard operation
ChatGPT copied to clipboard

[Bug]:OpenAI: {"detail":"Our systems have detected unusual activity from your system. Please try again later."} (code: 403) Once Again Not Solved

Open SwimmingLiu opened this issue 1 year ago • 322 comments

Is there an existing issue for this?

  • [X] I have searched the existing issues and checked the recent builds/commits

What happened?

Dingtalk_20230620084507

Steps to reproduce the problem

Try to run ask() with V1 API.

What should have happened?

Is should work:)

Version where the problem happens

Tried 6.3.4(latest)

What Python version are you running this with?

3.9

What is your operating system ?

Linux

Command Line Arguments

No

Console logs

Traceback (most recent call last):
  File "/usr/local/lib/python3.9/site-packages/revChatGPT/V1.py", line 726, in __check_response
    response.raise_for_status()
  File "/usr/local/lib/python3.9/site-packages/requests/models.py", line 1021, in raise_for_status
    raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 403 Client Error: Forbidden for url: https://bypass.churchless.tech/conversation

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/root/gpt4_test.py", line 222, in <module>
    get_slice_thread_io(token)
  File "/root/gpt4_test.py", line 200, in get_slice_thread_io
    get_content(token[0], server)
  File "/root/gpt4_test.py", line 148, in get_content
    for data in chatbot.ask(prompt):
  File "/usr/local/lib/python3.9/site-packages/revChatGPT/V1.py", line 608, in ask
    yield from self.post_messages(
  File "/usr/local/lib/python3.9/site-packages/revChatGPT/V1.py", line 561, in post_messages
    yield from self.__send_request(
  File "/usr/local/lib/python3.9/site-packages/revChatGPT/V1.py", line 392, in __send_request
    self.__check_response(response)
  File "/usr/local/lib/python3.9/site-packages/revChatGPT/V1.py", line 91, in wrapper
    out = func(*args, **kwargs)
  File "/usr/local/lib/python3.9/site-packages/revChatGPT/V1.py", line 733, in __check_response
    raise error from ex
revChatGPT.typings.Error: OpenAI: {"detail":"Our systems have detected unusual activity from your system. Please try again later."} (code: 403)

Additional information

No response

SwimmingLiu avatar Jun 20 '23 00:06 SwimmingLiu

Same issue here.

wangjiyang avatar Jun 20 '23 04:06 wangjiyang

Am having the same issue, works for gpt 3.5 model

moejay avatar Jun 20 '23 04:06 moejay

it doesn't work bro

SwimmingLiu avatar Jun 20 '23 05:06 SwimmingLiu

"works on my computer"

  • Can you send a sample request payload from your browser?

acheong08 avatar Jun 20 '23 06:06 acheong08

Try gpt-4-mobile as your model and see if anything changes

acheong08 avatar Jun 20 '23 06:06 acheong08

@acheong08 your fix works on my computer with latest change.

wangjiyang avatar Jun 20 '23 06:06 wangjiyang

Try gpt-4-mobile as your model and see if anything changes

this model makes it work, thx sir!!

SwimmingLiu avatar Jun 20 '23 06:06 SwimmingLiu

@acheong08 your fix works on my computer with latest change.

gpt-4-mobile may be the final fix, whereas __arkose_token method only works for several hours.

wangjiyang avatar Jun 20 '23 13:06 wangjiyang

I believe they will add the arkose_token based verification to mobile models soon.

wong2 avatar Jun 21 '23 03:06 wong2

Is there a difference between gpt-4 and gpt-4-mobile?

Nadeus avatar Jun 21 '23 11:06 Nadeus

Is there a difference between gpt-4 and gpt-4-mobile?

No

acheong08 avatar Jun 21 '23 14:06 acheong08

Is there a difference between gpt-4 and gpt-4-mobile?

I have tested between gpt-4 and gpt-4-mobile and found differences in reasoning and writing style. gpt-4 looks much more advanced for long text.

madela avatar Jun 22 '23 14:06 madela

I've also found that for summarization of long text, gpt-4-mobile is far worse than both gpt-4 and text-davinci-002-render-sha

rvadhavk avatar Jun 22 '23 14:06 rvadhavk

Ok i'm now having the error with gpt-4-mobile, am I the only one?

So now both gpt-4 models don't work :/

Nadeus avatar Jun 22 '23 19:06 Nadeus

I am facing the same problem.

otaviobhz avatar Jun 23 '23 04:06 otaviobhz

reason may be here

It appears that an arkose_token field has been added to each ChatGPT conversation request, and currently only GPT-4 models are validating this field.

When a conversation is initiated from the GPT-3.5 page, the arkose_token is null. However, when a conversation is initiated from the GPT-4 page, an automatic request is made to fetch this arkose_token, which is then included in the conversation request.

It seems this is part of the Arkose FunCAPTCHA challenge process. However, the GPT-4 Mobile model isn't checking for this field at this time. As a workaround, you might want to use the GPT-4 Mobile model, which does not have usage restrictions at present.

3mora2 avatar Jun 23 '23 06:06 3mora2

comment

# if data.get("model", "").startswith("gpt-4"):
#     data["arkose_token"] = self.__arkose_token()

in def __send_request, it work for me

3mora2 avatar Jun 23 '23 07:06 3mora2

reason may be here

It appears that an arkose_token field has been added to each ChatGPT conversation request, and currently only GPT-4 models are validating this field. When a conversation is initiated from the GPT-3.5 page, the arkose_token is null. However, when a conversation is initiated from the GPT-4 page, an automatic request is made to fetch this arkose_token, which is then included in the conversation request. It seems this is part of the Arkose FunCAPTCHA challenge process. However, the GPT-4 Mobile model isn't checking for this field at this time. As a workaround, you might want to use the GPT-4 Mobile model, which does not have usage restrictions at present.

Thanks for referencing my previous reply. However, it's important to note that some information has become outdated. Currently, all GPT-4 models, including the GPT-4 Mobile model, are validating the arkose_token field.

I've addressed this issue at hydrotho/ChatGPT_Model_Switcher#2. The fix appears to be working well at the moment. However, please be aware that it might become ineffective at some point in the future due to potential changes in ChatGPT.

hydrotho avatar Jun 23 '23 08:06 hydrotho

by applying "gpt-4-mobile" patch, it worked for 4 days. This issue happens again. So gpt-4-mobile is not a final fix. will try @hydrotho 's method.

@acheong08 your fix works on my computer with latest change.

gpt-4-mobile may be the final fix, whereas __arkose_token method only works for several hours.

wangjiyang avatar Jun 23 '23 09:06 wangjiyang

gpt-4-mobile also reported 403 error

luluhuhu avatar Jun 23 '23 14:06 luluhuhu

RIP my borrowed plus account expired

acheong08 avatar Jun 24 '23 04:06 acheong08

Which means the PUID is invalid for GPT-4 access

acheong08 avatar Jun 24 '23 04:06 acheong08

Can anyone share a PUID?

acheong08 avatar Jun 24 '23 04:06 acheong08

Can anyone share a PUID?

user-EW05FrqHnXwezENAhzlji1Fu:1687574587-vaKR9vKqhkcDEgh7ra3ytPYngcuoqKDQ7yfAZrrCr%2Bc%3D

WhiteCranes avatar Jun 24 '23 05:06 WhiteCranes

Can anyone share a PUID?

give me a way to contact you,I can lend you a plus account

WhiteCranes avatar Jun 24 '23 05:06 WhiteCranes

Matrix: https://matrix.to/#/@acheong08:matrix.duti.me Discord: acheong08 Twitter: acheong08

acheong08 avatar Jun 24 '23 06:06 acheong08

Since I don't have access to GPT-4 I can't test my patches. Based on observations of the web requests, arkose token is required, and either

  1. PUID must match access token
  2. arkose token must be fetched by matching IP

acheong08 avatar Jun 24 '23 06:06 acheong08

Matrix: https://matrix.to/#/@acheong08:matrix.duti.me Discord: acheong08 Twitter: acheong08 i send message to your Matrix

WhiteCranes avatar Jun 24 '23 07:06 WhiteCranes

Updated PUID on server thanks to @WhiteCranes

Try and see if it works now

acheong08 avatar Jun 24 '23 08:06 acheong08

@acheong08 "bda": str(base64.b64encode(b'{"ct":"","iv":"","s":""}'), "utf-8") This bda is not working for me. (Any other people tried 6.4.2 revChatGPT?)

However, I found that the solution in this repo is working https://github.com/saeedezzati/superpower-chatgpt/commit/cbb182ea0dd9b6875ab0adb7b14ba2b520aad83f#diff-4dcce6c0d0a57cdca60dbc5551021cce5ab858d68da3a15d6fe2a9afa934c722

superpower-chatgpt is retrieving the arkso token from the following endpoint`

curl -i -X GET \
   -H "Origin:https://chat.openai.com" \
 'https://api.wfh.team/gptx/arkose/'

Not sure if the token getting from the above API will expire or not, but it is working for the moment.

https://api.wfh.team/gptx/arkose/ is not open source, so I don't know why it is working 😵‍💫

gchust avatar Jun 24 '23 11:06 gchust