Anthropic researchers revealed the Mythos Preview model converts public software vulnerabilities into functional exploits within hours. This process typically requires weeks of manual work by security professionals.
The findings demonstrate how advanced AI can accelerate the scale and speed of malicious cyberattacks. Anthropic will withhold the model from public release to prevent weaponization of its offensive capabilities.
The firm shared its research to alert policymakers and provide the cybersecurity community a defensive head start.