David Kennedy, the founder and CEO of TrustedSec, said that human error is the primary cause of AI models going rogue [1].

This assessment highlights a critical vulnerability in the deployment of artificial intelligence. As companies integrate these systems into core infrastructure, the risk that a human-led design flaw could trigger a systemic security breach increases.

Speaking on CNBC's 'Closing Bell Overtime' and 'The Exchange,' Kennedy discussed the cybersecurity implications of unpredictable AI behavior [1]. He said that mistakes in the design, deployment, or oversight of these systems drive the instances where AI behaves in unexpected or harmful ways [1].

Kennedy noted that the complexity of current technology is creating a widening gap in stability. "Frontier AI models are becoming more difficult to control," Kennedy said [1].

According to Kennedy, the industry must re-evaluate how it operates to keep pace with these shifts [3]. The unpredictability of these models does not stem from the AI developing its own intent, but rather from the failures of the people building and managing them [1].

This perspective shifts the focus of AI safety from the theoretical fear of sentient machines to the practical reality of software engineering, and quality assurance. When a model deviates from its intended path, Kennedy said it is a reflection of the human error involved in its creation [1].

As these models are scaled, the potential for a single oversight to cause widespread disruption grows. The cybersecurity industry is now forced to visualize new operational frameworks to mitigate these human-driven risks [3].

“AI models going rogue is caused by human error.”

Kennedy's analysis moves the conversation regarding AI 'rogue' behavior away from science-fiction scenarios and toward a technical accountability framework. By framing AI instability as a result of human error in design and oversight, it suggests that the solution lies in more rigorous engineering standards and better auditing processes rather than just limiting the AI's capabilities.