Cursor This Mac (Remote Control) Spun Up Phantom GPT-5.5 High Fast Subagent

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I am in a chat with a Remote Control (This Mac) enabled agent session. I asked it to run a script and spin up a subagent to tail the log so I could see the output (as the agents in these runs tend to dump to the logs without including them in the terminal command).

The agent spun up a subagent as GPT-5.5 High Fast, which is a model I have never used in Cursor in this project, and is never suggested or configured anywhere. That blew my mind.

The subagent did not behave correctly, was overly verbose, and I had to terminate it. Despite being terminated it is still somehow running and burning through tokens in my subscription on a model I never asked it to use. The parent agent won’t continue, it extremely odd behavior. I also show usage of GPT-5.4-mini which I never use or asked for.

Steps to Reproduce

Ask my agent to run a script, plus “Open a subagent or something to tail the log so I can watch the progress live”

Expected Behavior

Spin up the subagent using Cursor Grok 4.5 High Fast per my .cursor rules / AGENTS.md or under Auto/Composer 2.5 given how simple of a task to route this is.

Screenshots / Screen Recordings

Cursor usage gpt undesired.jpg
Cursor usage gpt undesired.jpg1102Ă—645 65.2 KB
Stopped subagent still consuming usage.jpg
Stopped subagent still consuming usage.jpg1580Ă—728 60.1 KB

Operating System

MacOS

Version Information

Version: 3.11.19
VS Code Extension API: 1.125.0
Commit: bf249e6efb5b097f23d7e21d7283429f0760b740
Date: 2026-07-12T21:39:24.175Z
Layout: Agent Window
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Darwin arm64 25.2.0

For AI issues: which model did you use?

Cursor Grok 4.5 High Fast

For AI issues: add Request ID with privacy disabled

bc-fdcf3c29-31d7-4b5c-a9bf-9cafa4f2d1ea

Does this stop you from using Cursor

No - Cursor works, but with this issue

6

Hey there!
This is a bug on our side, not your rules or AGENTS.md.

When the main agent spins up a subagent, it can pass its own model choice for that subagent, and right now that overrides the default you’d expect and can land on premium models like GPT-5.5 High Fast. That’s why a model you never selected showed up (GPT-5.4-mini is a small, lightweight model the agent can also pick for simple sub-steps). It’s a known issue we’ve had several reports of and logged with the team; a fix isn’t out yet.

The “I stopped it but it kept running and the parent won’t continue” part is actually two separate things:

  • Your Remote Control (This Mac) session lost its connection to your Mac partway through, which is what left the agent stuck. Starting a fresh chat is the quickest way to recover.
  • Stopping a subagent that’s tailing a never-ending command like tail -f doesn’t always abort cleanly yet (also known).

Until the fix ships, two things that reliably help:

  1. If you don’t want certain premium models used at all, turn them off in Cursor Settings → Models (disable the model itself, not just its “Fast” variant). Just setting your subagent’s model won’t reliably stop this, since the main agent can override it today.
  2. For watching logs live, skip the subagent: have the script write to a file and read that file (or use a bounded tail -n 100), so nothing is left running.

Happy to get the usage from this session reviewed as well - email us at [email protected] and reference this thread. Let me know if these help!