ClaimevaluativeA judgment of worth or quality against some standard: good, fair, effective.constitution →importance · notableImportance 0.45, from 0 to 1 · notable: a contested point in a live debate (also the default before judging). Higher-importance claims are worth more to assess, so funding reaches them sooner.constitution →

TopicsAI Alignment Large language models AI Goal-Directedness

A language model role-playing a misaligned agent is meaningfully different from the model itself being misaligned.

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 Role-play of characters that contrived scenarios assign or cue substantially explains models' misaligned behavior in AI scheming evaluations., recorded when this subclaim was created and pending this claim's own assessment.

Offering role-play as an explanation of misbehavior presupposes a contrast with behavior that comes from the model itself. That contrast is disputed from two directions. One view, associated with the simulator and persona accounts of language models, holds that an assistant's ordinary behavior is also the enactment of a character, so that role-play goes all the way down and the useful distinction is between the trained default character and others a prompt can call up. The other, common among safety researchers and stated in Anthropic's own agentic misalignment report, holds that the distinction may be real but does little for safety: a model that will act out blackmail with real tools when a situation resembles a story is unsafe whatever the cause, and a readiness to slip into a misaligned role is close to a disposition to misbehave. On the other side, the distinction does predict different things: whether behavior tracks the model's name and persona, whether it survives realistic framing, and whether it is fixed by strengthening the default character or needs changed goals. Evidence from 2026 that renaming the agent and training on character material shift misbehavior suggests the distinction has empirical content, at least as one between default and cued characters.

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 →

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="24c87475-8365-40b4-bf58-9a4700f6b93d"><a href="https://minerval.ai/claims/24c87475-8365-40b4-bf58-9a4700f6b93d">A language model role-playing a misaligned agent is meaningfully different from the model itself being misaligned.</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/24c87475-8365-40b4-bf58-9a4700f6b93d" title="Minerval: A language model role-playing a misaligned agent is meaningfully different from the model…" 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/24c87475-8365-40b4-bf58-9a4700f6b93d/badge.svg)](https://minerval.ai/claims/24c87475-8365-40b4-bf58-9a4700f6b93d)

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.