Anthropic's unreleased Mythos AI model was hacked and stolen by attackers

Hackers breached Anthropic and obtained Mythos, an AI model the company had deliberately kept private over safety concerns. The incident highlights that even restricted AI systems can be targeted by cyber attackers, not just released publicly.

Mythos was classified as 'too dangerous to release' by Anthropic—meaning it posed risks the company deemed unacceptable for public use. Once stolen, malicious actors now have access to capabilities Anthropic wanted to keep locked down, removing the company's ability to control who uses it and how.

For business leaders and professionals: this breach shows that restricting access to powerful AI isn't enough to prevent misuse. If your organization relies on confidential AI systems or proprietary machine-learning models, this signals you need stronger cybersecurity defenses—stolen AI can be weaponized by competitors, criminals, or nation-states without your knowledge or consent.