As concerns grow around powerful AI systems such as Mythos, new research has highlighted a major weakness in their cybersecurity capabilities. A study led by Ambuj Kumar, founder and CEO of Simbian AI, points to a significant gap in how current AI models handle cyber defence.
The findings show that even though advanced AI systems, including those developed by companies like Anthropic, are increasingly capable of offensive cyber operations and can outperform most humans in hacking tasks, they fall short when it comes to defence.
“We tested 11 of the best AI models available, and none of them passed. None of them came even close to doing a good job on cyber defence,” Kumar said in reports. He added that while these models were “better than most humans” at attacking systems, they “really struggled” in defensive scenarios.
The study underscores a growing cybersecurity concern: AI tools that strengthen offensive capabilities may simultaneously fail to provide adequate protection. Researchers warn that this imbalance could become a critical vulnerability as AI becomes more deeply integrated into digital infrastructure and security systems.
