# Anthropic trains Claude to flag its own potential misuse after each reply

> Original title: Every time Claude answers, it now quietly asks itself whether it just helped someone cause harm, and that private verdict goes to Anthrop...

- Source: Rohan Paul
- Published: 2026-10-29T23:37:28.000Z
- Bad Influence: https://badinfluence.news/items/khcupf6ew0k4mpzmafno12cjz
- Original: https://x.com/rohanpaul_ai/status/2115951180748833146

## Summary

Anthropic says it has trained Claude to give a preliminary safety assessment of each conversation right after responding. The assessment flags misuse that its other safeguards were not designed to catch, and the company says it is efficient because Claude has already read the conversation. The author's post adds that the self-assessment matches 78% of a costlier classifier's flags on Fable 5 traffic, rising to 94% for high-priority cases, at a 1% false-alarm rate.

## Why it matters

The quoted post explains how Claude's self-assessment works and how Anthropic measured it against a costlier classifier, which is useful context for AI safety monitoring.

## Post

Every time Claude answers, it now quietly asks itself whether it just helped someone cause harm, and that private verdict goes to Anthropic’s reviewers.

Anthropic’s most accurate misuse detector is a Claude Mythos Preview classifier that reads all its safety policies before every ruling, which makes it too expensive to run on every chat. 

So Anthropic trained Claude to copy its judgments without the policies or the reasoning, and on Fable 5 traffic the copy matches 78% of the detector’s flags, rising to 94% for high-priority cases, at a 1% false-alarm rate.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHV1ZFPQbIAAPz0J.jpg&mode=thumb&exp=1793577600&sig=dd9dd5fa943dff4e)

## Quoted post @AnthropicAI

> We’ve trained Claude to flag potential misuse in conversations, a technique we call self-assessment. Right after it responds, Claude gives a preliminary safety assessment of the conversation. 
> 
> Since we began using self-assessment in June, it has surfaced misuse that our other safeguards weren’t designed to catch. 
> 
> Because Claude has already read the conversation, self-assessment is very efficient. And unlike classifiers, which target a discrete set of harms, it draws on Claude’s own judgment, so it gets more accurate as our models improve.
> 
> Read more on our new Safeguards blog:
> https://safeguards.anthropic.com/2026/self-assessment/

https://x.com/AnthropicAI/status/2115871889549865429