Why Anthropic’s new model has cybersecurity experts rattled

Q: Anthropic released Claude Mythos Preview, an AI model that found thousands of critical vulnerabiliti

Anthropic released Claude Mythos Preview, an AI model that found thousands of critical vulnerabilities in major operating systems including a 27-year-old OpenBSD flaw.

Q: Project Glasswing launched with 40+ tech giants including Apple, Google, and Microsoft to use Mythos

Project Glasswing launched with 40+ tech giants including Apple, Google, and Microsoft to use Mythos defensively before similar capabilities reach bad actors.

Q: Mythos achieved 93.9% on SWE-bench Verified versus 80.8% for the previous model, representing a 13%

Mythos achieved 93.9% on SWE-bench Verified versus 80.8% for the previous model, representing a 13% jump in AI capabilities since February.

Casey Newton·Tuesday, April 7, 2026·11 min read

AI/ML Technology Startups

Share𝕏 in

AI Summary

Anthropic announced Claude Mythos Preview, a new AI model that poses significant cybersecurity risks by finding thousands of vulnerabilities in major operating systems. The company launched Project Glasswing with 40+ tech companies to use the model defensively before similar capabilities proliferate to bad actors.

Key Facts

✓Anthropic released Claude Mythos Preview, an AI model that found thousands of critical vulnerabilities in major operating systems including a 27-year-old OpenBSD flaw.

✓Project Glasswing launched with 40+ tech giants including Apple, Google, and Microsoft to use Mythos defensively before similar capabilities reach bad actors.

✓Mythos achieved 93.9% on SWE-bench Verified versus 80.8% for the previous model, representing a 13% jump in AI capabilities since February.

Author Takes

BearishCasey Newton

AI regulation

Project Glasswing is built on the uncomfortable premise that the only way to protect us from dangerous AI is to build it first

BearishCasey Newton

AI centralization

This situation centralizes power and risk, with a private company now having powerful zero-day exploits

More from Casey Newton

The best argument I’ve heard for why AI won't take your job

Platformer launches a new podcast mini-series on AI and jobs, featuring Box CEO Aaron Levie arguing that AI will transform rather than eliminate most

May 13

The Trump administration's AI doomer moment

The Trump administration is reversing its anti-regulation stance on AI after Anthropic's new Mythos model demonstrated dangerous cybersecurity capabil

May 6

The week that Meta employees became training data

Meta has implemented invasive monitoring software called MCI on employee computers to capture their every mouse movement, click, and keystroke to trai

Apr 24