# AI models now surpass most humans at finding and exploiting software vulnerabilities, said Anthropic.

*genai · news · 2026-04-08 · Siliconrepublic.com*

## Key points

- Anthropic's unreleased Claude Mythos has found thousands of high-severity vulnerabilities in major systems.
- Mythos preview generated working exploits 181 times, while Opus 4.6 had near zero success.
- Over 40 critical infrastructure organizations now have access to Mythos preview for cyber defense.
- Anthropic committed up to $100 million in usage credits and $4 million to open-source security groups.
- Eirc Boyd, former Microsoft AI platforms president, has joined Anthropic as head of infrastructure.

**Companies:** Amazon Web Services, Apple, Google, Microsoft, Broadcom, Cisco, CrowdStrike, JP Morgan Chase, Nvidia, Palo Alto Networks
**Countries:** United States

[Read the full story on Siliconrepublic.com](https://www.siliconrepublic.com/enterprise/anthropics-glasswing-project-employs-mythos-to-prevent-ai-cyberattacks)

---

Canonical: https://newsio.io/n/e601d6ff-becb-43cb-af91-dde90e2eeb1a/ai-models-now-surpass-most-humans-at-finding-and-exploiting-software-vulnerabili
Summarized by Newsio from Siliconrepublic.com. https://newsio.io/how-it-works