A comprehensive study by the Center for AI Safety has identified significant limitations in current alignment techniques when applied to more advanced AI architectures. The research demonstrates that methods such as RLHF (Reinforcement Learning from Human Feedback) may not scale reliably to systems with substantially greater capabilities. The study proposes that more theoretical work is needed to develop alignment approaches with provable guarantees. Researchers also highlight the importance of complementary measures such as rigorous testing protocols and governance frameworks rather than relying solely on technical alignment solutions.