Anthropic has launched Claude Mythos Preview, a cybersecurity AI model with advanced vulnerability detection capabilities, but is limiting access to select organizations due to dual-use concerns. The restricted release follows two recent data leaks that exposed internal documents and source code, raising questions about Anthropic's security practices. The model can identify vulnerabilities at unprecedented scale but could also be weaponized by malicious actors if broadly available.
Background
Anthropic is an AI safety company known for developing Claude AI models with constitutional AI principles. The cybersecurity AI space has seen growing interest in using large language models for threat detection and analysis.
- Source
- Ars Technica
- Published
- Apr 8, 2026 at 09:34 PM
- Score
- 7.0 / 10