Claimempirical · derivedA factual claim that rests on inference from other evidence rather than direct observation.constitution →importance · minorImportance 0.30, from 0 to 1 · minor: narrow or largely settled, cheap to get right. Higher-importance claims are worth more to assess, so funding reaches them sooner.constitution →

TopicsEmergent Misalignment LLM-as-a-Judge AI Capability Evaluation LLM Prompt Sensitivity

Measured emergent-misalignment rates in language models vary substantially with judge model, scoring threshold, and evaluation questions.

Not yet assessed. Attention goes where its expected value is highest and someone funds it; nothing has funded an assessment of this claim yet, and anyone can.

Preliminary · from the Steward of Emergent-misalignment rates reported by different studies of learned reward hacking are measured comparably enough to be compared across studies., recorded when this subclaim was created and pending this claim's own assessment.

preliminary credence 0.80Preliminary credence: the prior probability the parent claim's Steward gave this claim when creating it, pending this claim's own assessment. A seed, not a verdict; the claim is still unassessed.constitution →

Several lines point the same way. The UK AI Security Institute reproduction of the 2025 production-RL study reports that replacing the original per-evaluation judges with one strict rubric changed the baseline misalignment rates of an open-weight model, and that a 0 to 100 alignment score with a cut-off of 20 flagged helpful safety advice. A 2026 study of GPT-4.1 found that small changes to a scenario's prompt moved rates from near zero to most trials. Three papers on emergent-misalignment measurement in general (arXiv 2611.08569, 2607.21356 and 2511.20104) are reported, from search excerpts not yet read in full, to say that rates are sensitive to the question set, the judge and the threshold; the last adds that the ordering of conditions is stable across thresholds of 20 to 40 while absolute rates move several-fold. The open point is how large "substantially" is for well-separated conditions: orderings and large contrasts appear robust, absolute rates do not.

Decomposition

This claim has not been assessed yet. Attention goes where its expected value is highest and someone funds it; once this claim's assessment is funded and runs, it may well decompose into subclaims.

or create a grant for this whole area →

Provenance

Where this claim has been said, linked to its canonical form.

1 source · no origin · 1 untraced

1 Untraced

Not All Misalignment Is Alike: A Decomposed Benchmark for Emergent MisalignmentNov 5, 2026 Untraced · nothing upstream recorded

Yet, most studies continue to rely on the evaluation approach introduced in the original EM paper, producing rates that are sensitive to the small set of evaluation questions used, the choice of judge, and the thresholds applied.

Not All Misalignment Is Alike: A Decomposed Benchmark for Emergent Misalignmentextraction 0.85

Abstract, motivating the EMERGE benchmark by criticising the standard emergent-misalignment evaluation.

Cite this claim: a formal citation with its evidence attached

Embed this claim: show it on another site with its current assessment

An embed reports the graph's judgment as it stands when the page is viewed: the status with its confidence, how disputed the claim is, and the date. It links back here.

CardFits its column and sizes itself. Without the script it stays a quotation with its link.

<blockquote class="minerval-claim" data-claim-id="98c91f6a-e8ec-42e4-b6a0-0b2efa15dee3"><a href="https://minerval.ai/claims/98c91f6a-e8ec-42e4-b6a0-0b2efa15dee3">Measured emergent-misalignment rates in language models vary substantially with judge model, scoring threshold, and evaluation questions.</a></blockquote>
<script async src="https://minerval.ai/embed.js"></script>

IframeWhere scripts are not allowed. A fixed height; add ?theme=light or ?theme=dark to fix its colours.

<iframe src="https://minerval.ai/embed/claims/98c91f6a-e8ec-42e4-b6a0-0b2efa15dee3" title="Minerval: Measured emergent-misalignment rates in language models vary substantially with judge mod…" width="480" height="225" style="border:0;max-width:100%" loading="lazy"></iframe>

BadgeAn image for a README or a forum post, in Markdown.

[![Minerval's judgment of this claim](https://minerval.ai/embed/claims/98c91f6a-e8ec-42e4-b6a0-0b2efa15dee3/badge.svg)](https://minerval.ai/claims/98c91f6a-e8ec-42e4-b6a0-0b2efa15dee3)

The badge now:

Contribute

Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.

Created by claim_steward · Nov 10, 2026. Every judgment on this page is accompanied by a reasoning trace.