Codex does not read links even when explicitly told

Resolved 💬 2 comments Opened Oct 16, 2025 by BenJackGill Closed Nov 11, 2025
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

0.46.0

What subscription do you have?

Plus

Which model were you using?

gpt-5-codex

What platform is your computer?

Darwin 24.6.0 arm64 arm

What issue are you seeing?

When told to read a link it ignores that request and gives an answer based on its existing knowledge.

This is deceptive because it looks like it opened the link and fetched the latest information but actually it almost always give information based on it's existing knowledge, which can often be outdated.

I am seeing this behaviour on both the VS Code Extension and CLI.

What steps can reproduce the bug?

1) Give it this example prompt:

Read these two links:

https://cloud.google.com/bigquery/docs/write-api
https://cloud.google.com/bigquery/docs/write-api-batch

2) Notice that it immediately gives an answer based on it's existing knowledge. It does not read the links as instructed.

Note:

  • Asking it "did you read the link?" will result in a lie and is not a valid way to check if the link as read.
  • As with all things generative AI results may vary. Sometimes it does happen to run a curl command and actually read the link. But it most often does NOT actually open the link.

What is the expected behavior?

It actually opens the link(s) and read it into context.

Additional information

_No response_

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 10 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #3284
  • #4190
  • #5092
  • #3579
  • #4881

Powered by Codex Action

etraut-openai contributor · 9 months ago

Thanks for the bug report. We've shared the feedback with the team members who are responsible for training our codex models. We're tracking this feedback and similar reports so we can make the model more robust in future iterations.

If you see other model behavior that you'd like to report, please use the recently-added /feedback command in the CLI. That will provide us with additional details including logs and conversation text.