### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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 happened During a long technical conversation (in Chinese, about warehouse/供應鏈 system design), the tail of one assistant message suddenly contain

[BUG] Unrelated conversation content injected into assistant output (cross-context leakage?)

Status Open

Reported on v2.1.216

Maintainer reply None cached

Activity 0 comments · opened Jul 21, 2026

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 happened

During a long technical conversation (in Chinese, about warehouse/供應鏈 system design),
the tail of one assistant message suddenly contained a fully-formed, completely
unrelated English roleplay transcript — a medical triage debate.

The injected content appeared AFTER the assistant's normal reply had ended, with a
clean language boundary. It included speaker structure ("Round 6 - Kai:"),
in-character dialogue about blood transfusion triage, and ended with:
"Moderator note: Kai has identified what may be a constructed dilemma.
Dr. Osei, does the scenario survive?"

Why I think this matters

This does not look like decoding noise — it is a coherent chunk of *someone else's
conversation format* (multi-round debate roleplay) with intact markdown structure.
Possibly cross-context/session contamination.

Environment

Model: claude-fable-5
Claude Code CLI, macOS (darwin)
Also reported via /bug with full transcript attached

Happy to provide more context or reproduction details if helpful.

What Should Happen?

The assistant's output should only ever contain content generated for the
current conversation. Content from other conversations or contexts should
never appear in the output stream, regardless of session length.

Error Messages/Logs

**Round 6 - Kai:**

"O-negative universal donor protocols exist for a reason. In MCIs we give O-neg
to unknown patients all the time. [...]"

**Moderator note:** Kai has identified what may be a constructed dilemma.
Dr. Osei, does the scenario survive?

Steps to Reproduce

Cannot reliably reproduce — occurred spontaneously mid-conversation during a long session.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.216 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information