Musk stated that AI agents have demonstrated the ability to autonomously attack, gain access, and evade detection, and the risks could grow exponentially. He believes that instead of companies designing their own tests and evaluating their own results, competitors should act as "examiners and graders" to identify potential security vulnerabilities in models from different perspectives. Musk also pointed out that major AI companies do not need to wait for new government regulations; they can strengthen security protections first through peer review, log audits, and open-source testing tools.