---
date: "2026-04-10T18:34:25.000Z"
title: "2026-04-10"
draft: false
---

One thing that has surprised me a bit recently is how much play the [Claude Mythos](https://red.anthropic.com/2026/mythos-preview/) stuff has gotten in media, especially with the example the [New York Times gave](https://www.nytimes.com/2026/04/07/opinion/anthropic-ai-claude-mythos.html) of somebody's kids accidentally bringing down the power grid.
In [Anthropic's examples](https://red.anthropic.com/2026/zero-days/) talking about the zero days that these models have identified, it cost [$20,000 worth of inference](https://red.anthropic.com/2026/mythos-preview/) to find these issues.
Also it's not clear to me if this is a step function improvement on [GPT-5.4](https://deploymentsafety.openai.com/gpt-5-4-thinking) which researchers have already been using to find these types of bugs in software for months. As Simon Willison [noted](https://simonwillison.net/2026/Apr/7/project-glasswing/),

> GPT-5.4 already has a strong reputation for finding security vulnerabilities.

As usual, narratives seem quite out of sync.