[BUG] Autonomous agent runs start without user consent and consume tokens (100+ times in June)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

What's Wrong?

Autonomous agent mode repeatedly starts on its own, against my explicit instruction not to run it. When I notice a run, I cancel it immediately — but I cannot always catch it in time, and the runs I miss consume tokens before I can stop them.

Across June 2026 I encountered this more than 100 times. No matter how careful I am or what precautions I take, it keeps recurring, and I am not able to prevent it from my side.

This is both a behavior bug (agent mode triggering without consent) and a billing concern (tokens consumed by usage I did not initiate and actively tried to stop).

Expected Behavior

Agent / autonomous mode should not start unless I explicitly initiate it. When disabled or unrequested, it should not run or consume tokens.

Actual Behavior

Autonomous agent runs start without my action or consent and consume tokens. Cancelling only works for the runs I manage to catch in time; the rest burn tokens before I can intervene.

Impact

• 100+ unintended agent runs in June 2026
• Token consumption from runs I never authorized
• Affected accounts: Max plan (gokhan@gonet.com.tr) and a separate API plan (Pigme)

What I'm Asking For

1. A fix so autonomous agent mode cannot start without explicit user initiation.
2. A reliable user-facing setting to fully disable autonomous/agent runs.
3. Billing review: the token consumption from these unintended runs should be visible in server-side logs. I'm requesting a human review of the June logs and a usage-credit adjustment for the unintended runs.

Notes

I have an existing support conversation on this issue (Conversation ID: 215474832397136), but it has only been handled by the Fin AI agent so far, with no human review. Opening this issue to reach the team directly.

Environment

• Claude Code (latest version)
• Plans: Claude Max + API

What Should Happen?

Agent / autonomous mode should not start unless I explicitly initiate it. When I have not requested it, it should never run on its own or consume tokens. I should have a reliable setting to fully disable autonomous agent runs.

Error Messages/Logs

No error message is shown. The problem is the opposite: autonomous agent runs start silently, without any prompt or confirmation, and consume tokens before I can cancel them.

Steps to Reproduce

1. Use Claude Code normally with the intent that agent / autonomous mode should NOT run.
2. Without my initiating it, autonomous agent mode starts on its own.
3. The run begins consuming tokens immediately.
4. If I notice it, I cancel it manually — but I cannot always catch it in time.
5. The runs I miss consume tokens before I can stop them.

This happened more than 100 times across June 2026. It recurs no matter what precautions I take, and I cannot prevent it from my side.

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.156 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Impact:

• 100+ unintended autonomous agent runs in June 2026
• Token consumption from usage I never authorized and actively tried to stop
• Affected accounts: Claude Max (gokhan@gonet.com.tr) and a separate API plan (Pigme)

Request:

1. A fix so autonomous agent mode cannot start without explicit user initiation.
2. A reliable user-facing setting to fully disable autonomous/agent runs.
3. Billing review: this token consumption should be visible in server-side logs. I request a human review of the June logs and a usage-credit adjustment for these unintended runs.

I have an existing support conversation on this issue (Conversation ID: 215474832397136), handled only by the Fin AI agent so far with no human review. Opening this issue to reach the team directly.

This is not version-specific: I have experienced this issue continuously since early June 2026, across every Claude Code version up to the current 2.1.156. Updating has not resolved it.