Vision model usage in disabled organization breaks the agent session

[mafshin]

If the model generates a tool call which requires vision then there's no way to get that session back to work. I lost several premium requests just retrying without any progress.

• Copilot Chat Extension Version: 0.40.1
• VS Code Version: 1.112.0
• OS Version: Win11
• Feature (e.g. agent/edit/ask mode): agent
• Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Opus 4.6

Steps to Reproduce:

1. Make the model to read an image when vision is disabled for the organization e.g. using playwright cli
2. Request will fail and there's no way to continue on that session anymore

[justschen]

super strange - images shouldn't even be sent in the first place so we're probably missing a policy check somewhere! will give this a look, thanks

[zzefff]

Same issue here

VSCode Version: 1.113.0 (Universal)
copilot-chat extension Version: 0.41.1

[patelriki13]

Any updates on this issue?

Could someone confirm whether this is a bug or expected behavior?

I’m experiencing the same problem. When using the Claude Opus 4.6 model with images added in the context folder, it fails to process them. However, the same setup works correctly with GPT-5.3-Codex, which is able to read and interpret the images.

I also couldn’t find any configuration related to vision support. If this is a limitation, could you clarify why it works with one model but not the other?

VSCode Version: 1.113.0
copilot-chat extension Version: 0.41.2

[rjgoyal26]

I am stuck in the same loop in the my chat window. Yet, I am unsure if this is a company policy "feature/bug" or if it is a VSC bug at this point... using also the latest VSC 1.113.0 with Claude Opus 4.6

[mafshin]

This is blocking several of my sessions and with Opus we are losing several premium requests., please make it a high priority. Reproduced on 1.113.0.