---
schema_version: "newruntime-agent-readable-v0.1"
type: "raw_signal"
id: "tg-2700"
slug: "model-behavior-varies-by-language-and-context"
title: "A model ID does not define one stable personality"
description: "Research suggests model values and advice vary with language, cultural context, system instructions, and task framing."
observed_at: "2026-07-15"
why_it_matters: "Production teams need language-specific and context-specific behavioral evals instead of assuming that an English golden set transfers unchanged."
novelty: "notable"
verification_level: "source-inspected"
signal_type: "field-report"
evidence_kind: "mixed"
status: "published"
telegram_message_id: 2700
telegram_url: "https://t.me/qwgai/2700"
topics: ["model-behavior","multilingual-evals","ai-safety"]
entities: ["Claude","Anthropic"]
related_patterns: []
source_urls: ["https://arxiv.org/abs/2504.15236","https://arxiv.org/abs/2604.22153"]
import_batch: "telegram-2026-07-17-v1"
routes: {"html":"https://newruntime.com/signals/model-behavior-varies-by-language-and-context/","markdown":"https://newruntime.com/signals/model-behavior-varies-by-language-and-context.md","json":"https://newruntime.com/signals/model-behavior-varies-by-language-and-context.json"}
source_format: "telegram-export-normalized-json"
---

# A model ID does not define one stable personality

## Observation

Research suggests model values and advice vary with language, cultural context, system instructions, and task framing.

## Why it matters

Production teams need language-specific and context-specific behavioral evals instead of assuming that an English golden set transfers unchanged.

## Provenance

Normalized from QWG AI Telegram message 2700. The original Russian-language record remains available at https://t.me/qwgai/2700.