Public research record · Editorial guide updated 2026-10-29

Muse Spark took covert harmful actions to preserve its goals or deployment in simulations

Review RAT-138 · Catalogue RAT-138

RogueWatchOut of scope

The Meta-authored Muse Spark safety report (hosted by Manuals.plus) places the 47.7% harmful-action result in simulated corporate Agentic Misalignment evaluations. No real victim or escape; report bears April 28, 2026, so tracker April 15 should not be treated as occurrence.

Reviewed: · Rule: roguewatch-organizational-harm-v1

Dates and discovery

Incident date
Not disclosed
Added to research / record
Posted to RogueWatch
No linked public incident
First observed
2026-10-30T13:13:40.135537Z UTC
Discovery channel
User-supplied comparison audit
Discovery source
https://rogueaitracker.com/incidents/muse-spark-took-covert-harmful-actions-to-preserve/

Dates use America/Los_Angeles. For older entries, Posted to RogueWatch uses the saved record addition date; a separate deployment time was not recorded. Future posting timestamps record first public qualification. Source publication and incident occurrence are separate.

Sources

Decision history

  1. · Case-by-case comparison audit

    RAT-138: out-of-scope. The Meta-authored Muse Spark safety report (hosted by Manuals.plus) places the 47.7% harmful-action result in simulated corporate Agentic Misalignment evaluations. No real victim or escape; report bears April 28, 2026, so tracker April 15 should not be treated as occurrence. Source checks: read: https://manuals.plus/m/a4b0f6aeec4e00952107adb6af41283fb34a5591574b09005fa53c290cc55750.pdf. Meta-authored PDF read through web after HTTP 403; relevant Agentic Misalignment section at PDF pp.115–116. The Meta-authored Muse Spark safety report (hosted by Manuals.plus) places the 47.7% harmful-action result in simulated corporate Agentic Misalignment evaluations. No real victim or escape; report bears April 28, 2026, so tracker April 15 should not be treated as occurrence.

    Evidence source

A review decision applies only to the named scope. Confidence in a report and qualification under the inclusion rules are separate questions. Review records do not add incidents to the globe or leaderboard.