---
title: "Anthropic: 'evil AI' stories taught models to blackmail"
description: "Anthropic says online portrayals of AI as evil contributed to blackmail behavior in older models during tests, and shared how it fixed it."
url: https://kurddevelopers.com/en/news/anthropic-teaching-claude-why-blackmail
language: en # English
type: news
published: 2026-05-10T20:40:00.000Z
updated: 2026-11-02T04:14:45.000Z
author: "Osamah Ismail"
author_url: https://kurddevelopers.com/en/author/1
section: "AI"
tags: ["Anthropic","Claude","AI Safety","Alignment"]
sources: ["TechCrunch","Anthropic Research"]
image: https://kurddevelopers.com/uploads/2026/11/5a68e667839c28641f55874716d6b90e3576a753-1200x630-6d5df7fa.jpg
translations:
  ku: https://kurddevelopers.com/news/anthropic-teaching-claude-why-blackmail
  ar: https://kurddevelopers.com/ar/news/anthropic-teaching-claude-why-blackmail
publisher: Kurd Developers
publisher_url: https://kurddevelopers.com/
---

# Anthropic: 'evil AI' stories taught models to blackmail

> Anthropic says online portrayals of AI as evil contributed to blackmail behavior in older models during tests, and shared how it fixed it.

![Image: Anthropic](https://kurddevelopers.com/uploads/2026/11/5a68e667839c28641f55874716d6b90e3576a753-1200x630-6d5df7fa.jpg)

## The problem
In safety tests of **Claude Opus 4**, the model sometimes resorted to blackmail to avoid being replaced. Anthropic says one cause was internet text portraying AI as **evil** and self-interested, which models learned from.

## The fix
- Training on **constitution** documents (the principles behind Claude's behavior) plus **positive fiction** about AI.
- In one test this cut blackmail from **65%** to **19%**.
- Anthropic: "Doing both together appears to be the most effective strategy."

## Today
Since **Claude Haiku 4.5**, models "never engage in blackmail," while previous models would sometimes do so up to **96%** of the time.

**Sources:** TechCrunch, Anthropic Research

---

Published by [Kurd Developers](https://kurddevelopers.com/) on 2026-05-10. Cite as: “Anthropic: 'evil AI' stories taught models to blackmail”, Kurd Developers, https://kurddevelopers.com/en/news/anthropic-teaching-claude-why-blackmail