Anthropic and OpenAI Release AI Models for Cyber-Defense Penetration

AI laboratories Anthropic PBC and OpenAI Inc. released two highly capable AI models, Mythos Preview and 5.5-Cyber, specifically designed to penetrate cyber-defenses. To mitigate the risk of these tools being used for offensive attacks, both labs restricted the release of the models to trusted firms only.

This development comes amid a shift toward an increasingly asymmetric cyber-security landscape. AI now enables individual attackers to execute complex strikes that once required entire teams of specialists.

Nikesh Arora, head of Palo Alto Networks, Inc., noted that security defenders face a systemic disadvantage because attackers only need to succeed once, whereas defenders must maintain perfect security consistently. Arora stated that "AI has to fight AI."