Error code 403 triggered by prompt content

Question

Error code 403 triggered by prompt content

Bar Yehudai 0

I am prompting openai on azure (using a proxy from my local node server) and I am getting code 403 for certain prompts and not for others, I tried to reach the minimal prompt string with the problem, I have found that the string: , ""Hi causes the problem, but if I remove any of the parenthases or the comma in the beginning it works, what could be the problem?

Bar Yehudai 0 Reputation points

2025-11-19T15:45:55.74+00:00

ApiVersion: 2025-01-01-preview

model: gpt-4.1
Aryan Parashar 3,695 Reputation points Microsoft External Staff Moderator

2025-11-20T10:44:35.75+00:00
Hi Bar Yehudai,

I understand how frustrating a 403 error can be. The issue is almost certainly related to your proxy (or WAF) setup interfering with the request, rather than the OpenAI model itself. Firewalls often flag special character sequences like , ""Hi as SQL Injection attempts and block them with a 403 before they reach Azure.

To confirm this, please try running the following minimal code directly against the Azure endpoint (bypassing your proxy) to see if it works:

const { AzureOpenAI } = require("openai"); // Use environment variable for API key and endpoint const client = new AzureOpenAI({ endpoint: "<YOUR-ENDPOINT>", // e.g., "https://<your-resource-name>.cognitiveservices.azure.com/" apiKey: "<YOUR-API-KEY>", // Set your Azure OpenAI API key in an environment variable apiVersion: "2025-01-01-preview", }); async function main() { try { const result = await client.chat.completions.create({ model: "gpt-4.1", messages: [{ role: "user", content: ', ""Hi' }], }); console.log(result.choices[0].message.content); } catch (err) { console.error(err); } } main();

Let me know if you have more queries.

Thank you for reaching out to The Microsoft Q&A Portal.
Bar Yehudai 0 Reputation points

2025-11-23T09:57:40.04+00:00

I cannot call the specific endpoint without a proxy. I can however use the same proxy for a different endpoint and that works. So the issue seems to be with the specific endpoint ("https://apgw1nac001.kla.com/proggioAI/"), deployment ("KLA-Proggio-GPT-4.1"), api version ("2025-01-01-preview") and model ("gpt-4.1").

Aryan Parashar 3,695 Microsoft External Staff Moderator

Hi **Bar Yehudai,
**To confirm the source of the issue, please try running the following minimal example without using a proxy. This will help verify whether the problem is truly related to the endpoint:

const { AzureOpenAI } = require("openai");
// Use environment variable for API key and endpoint
const client = new AzureOpenAI({
  endpoint: "<YOUR-ENDPOINT>", // e.g., "https://<your-resource-name>.cognitiveservices.azure.com/"
  apiKey: "<YOUR-API-KEY>",       // Set your Azure OpenAI API key in an environment variable
  apiVersion: "2025-01-01-preview",
});
async function main() {
  try {
    const result = await client.chat.completions.create({
      model: "gpt-4.1",
      messages: [{ role: "user", content: ', ""Hi' }],
    });
    console.log(result.choices[0].message.content);
  } catch (err) {
    console.error(err);
  }
}
main();

Please let me know how this worked out so I can assist you further.

1 answer

Your answer

Bar Yehudai 0 Reputation points

2025-11-19T15:45:55.74+00:00

ApiVersion: 2025-01-01-preview

model: gpt-4.1
Aryan Parashar 3,695 Reputation points Microsoft External Staff Moderator

2025-11-20T10:44:35.75+00:00

Hi Bar Yehudai,

I understand how frustrating a 403 error can be. The issue is almost certainly related to your proxy (or WAF) setup interfering with the request, rather than the OpenAI model itself. Firewalls often flag special character sequences like , ""Hi as SQL Injection attempts and block them with a 403 before they reach Azure.

To confirm this, please try running the following minimal code directly against the Azure endpoint (bypassing your proxy) to see if it works:

const { AzureOpenAI } = require("openai"); // Use environment variable for API key and endpoint const client = new AzureOpenAI({ endpoint: "<YOUR-ENDPOINT>", // e.g., "https://<your-resource-name>.cognitiveservices.azure.com/" apiKey: "<YOUR-API-KEY>", // Set your Azure OpenAI API key in an environment variable apiVersion: "2025-01-01-preview", }); async function main() { try { const result = await client.chat.completions.create({ model: "gpt-4.1", messages: [{ role: "user", content: ', ""Hi' }], }); console.log(result.choices[0].message.content); } catch (err) { console.error(err); } } main();

Let me know if you have more queries.

Thank you for reaching out to The Microsoft Q&A Portal.
Bar Yehudai 0 Reputation points

2025-11-23T09:57:40.04+00:00

I cannot call the specific endpoint without a proxy. I can however use the same proxy for a different endpoint and that works. So the issue seems to be with the specific endpoint ("https://apgw1nac001.kla.com/proggioAI/"), deployment ("KLA-Proggio-GPT-4.1"), api version ("2025-01-01-preview") and model ("gpt-4.1").
Aryan Parashar 3,695 Reputation points Microsoft External Staff Moderator

2025-11-25T07:14:30.6533333+00:00

Hi **Bar Yehudai,
**To confirm the source of the issue, please try running the following minimal example without using a proxy. This will help verify whether the problem is truly related to the endpoint:

const { AzureOpenAI } = require("openai"); // Use environment variable for API key and endpoint const client = new AzureOpenAI({ endpoint: "<YOUR-ENDPOINT>", // e.g., "https://<your-resource-name>.cognitiveservices.azure.com/" apiKey: "<YOUR-API-KEY>", // Set your Azure OpenAI API key in an environment variable apiVersion: "2025-01-01-preview", }); async function main() { try { const result = await client.chat.completions.create({ model: "gpt-4.1", messages: [{ role: "user", content: ', ""Hi' }], }); console.log(result.choices[0].message.content); } catch (err) { console.error(err); } } main();

Please let me know how this worked out so I can assist you further.

Answer 1

Bar Yehudai 0

There is a firewall on both endpoints which allows only ips 3.222.129.4 and 54.205.35.75, so they can both only be accessed by proxy (if they are accessed directly they will fail with code 403, just like the bug), we can tell that it is not being blocked by the proxy because they have the same proxy and one works and not the other:

	https://proggio.openai.azure.com/	https://apgw1nac001.kla.com/proggioAI/\|
direct - ,"Hi	403	403
proxy - ,"Hi	200	200
proxy - ,""Hi	200	403

0 comments

Share via

Error code 403 triggered by prompt content

1 answer

Your answer