Claude (Sonnet 4.6) on AUTOPILOT seems wrongly integrated, doesnt terminate, stop or do anything, keeps infinate internal conversation

### Describe the bug

Since this was working a month ago something fundamentally seem to have been broken.

When select the mentioned model then nothing results from it, it just keeps infinitely looking on internal conversation, this wasnt the case when i used copilot last month, so something must have been broken in the Anthropic model integration.

### Affected version

GitHub Copilot CLI 1.0.40

### Steps to reproduce the behavior

1. Select Sonnet 4.6 (maybe also other, i only tested this)
2. set on autopilot maybe
3. ask it to do literally anything
4. wont reach a result

### Expected behavior

just reach a conclusion, result, fix whatever, instead of infinitely circling in internal conversation and draining tokens

### Additional context

$PSVersionTable

Name Value
---- -----
PSVersion 7.6.1
PSEdition Core
GitCommitId 7.6.1
OS Microsoft Windows 10.0.26200
Platform Win32NT
PSCompatibleVersions {1.0, 2.0, 3.0, 4.0…}
PSRemotingProtocolVersion 2.4
SerializationVersion 1.1.0.1
WSManStackVersion 3.0

Start by reproducing the issue in GitHub Copilot CLI 1.0.40 with Claude Sonnet 4.6 selected in autopilot on PowerShell 7.6.1 under Windows 10. Compare the behavior with autopilot disabled or another model, and consider the issue resolved when the CLI reaches a conclusion or stops instead of looping indefinitely and draining tokens.

Written by the indexing model from the issue text.