The issue reported in GitHub issue #74066 for `claude-code` describes a potential session or cache leakage where a user, while working in an Enterprise ZDR workspace, observed the AI agent referencing content ("building a Minecraft temple") that was unrelated to their current task. ### Summary of the Situation: * **The Incident:** The user reported that their `claude-code` session suddenly began discussing a "Minecraft temple," which they suspected was either leaked from a colleague's session within the same enterprise environment or, more concerningly, leaked from a separate consumer-plan session. * **Why it Matters:** If confirmed, this represents a significant security and privacy vulnerability. It suggests that session context, cache, or prompt history might not be properly isolated between different workspace instances or user accounts. For enterprise users, this raises concerns about the potential exposure of sensitive proprietary data or internal chat sessions. * **Current Status:** The issue was opened on July 4, 2026, and is categorized under `area:core` and `area:security`. As of now, it is a reported bug and is being investigated. ### Source: * **Primary Source:** [GitHub Issue #74066: [Bug] Potential session/cache leakage between workspace instances or consumer accounts](https://github.com/anthropics/claude-code/issues/74066) There is currently no broader public confirmation or widespread reporting of this specific leakage issue beyond this individual bug report. The investigation is ongoing within the `claude-code` repository.