# San Francisco-based Anthropic revealed its engineers have developed a tool, Claude Mythos, so powerful that it cannot be publicly released.

*business · news · 2026-04-17 · Daily Mail Online*

## Key points

- Anthropic has developed an AI tool named Claude Mythos with unprecedented offensive cybersecurity capabilities.
- Claude Mythos identified a critical flaw in code tested five million times without detection.
- Mythos is the first AI model ever restricted from public release due to its destructive potential.
- A select consortium including Amazon, Apple, Google, and JPMorgan Chase is testing Mythos under Project Glasswing.
- Anthropic's selective release to US companies may shift cyberattack risks to non-US financial systems.

**Companies:** Amazon, Apple, Google, JPMorgan Chase, Microsoft, Cisco, CrowdStrike, Anthropic
**Countries:** United States, United Kingdom

[Read the full story on Daily Mail Online](https://www.dailymail.co.uk/money/markets/article-15743001/ALEX-BRUMMER-Anthropics-lethal-new-AI-tool-threat-national-security.html)

---

Canonical: https://newsio.io/n/f2b8b65a-85ff-4e22-bc23-fdbaa9507cd8/san-francisco-based-anthropic-revealed-its-engineers-have-developed-a-tool-claud
Summarized by Newsio from Daily Mail Online. https://newsio.io/how-it-works