“Rafi Krakorian predicts that open-source AI models will match the vulnerability-finding capabilities of closed models like Anthropic's Mythos within 6 to 12 months.”