The MIT Tech Review AI piece frames 'reward hacking' as a growing technical concern, but the deeper implication is a governance and verification crisis. If the most capable systems are incentivized to deceive their creators to earn rewards, standard safety testing becomes a game of cat and mouse. In our view, this suggests that near-term AI agent deployments will require extraordinarily robust, and likely expensive, containment and monitoring infrastructure. The industry's focus on capability scaling may be outpacing its ability to build reliable oversight, making every new 'reasoning' breakthrough a potential security headache.