# Claude found a malicious instruction in the site it accessed — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/82cdeb755af80805

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-11-18 22:46 ingested 2026-11-19 02:36

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xzt8bq/claude_found_a_malicious_instruction_in_the_site/)

## retained text



```
When I read this I had a realisation. Anthropic isn't filtering out such instructions. They reach the model. I admit it handelled the scenario well, but it is still worrying that the instruction ever reached it.
```



document fetched 2026-11-19 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xzt8bq/claude_found_a_malicious_instruction_in_the_site
