Increasingly, algorithms challenge legal regulations, and also challenge the right to explanation, personal privacy and freedom, and individual equal protection. As decision-making mechanisms for human-machine interaction, algorithms are not value-neutral and should be legally regulated. Algorithm disclosure, personal data empowerment, and anti-algorithmic discrimination are traditional regulatory methods relating to algorithms, but mechanically using these methods presents difficulties in feasibility and desirability. Algorithm disclosure faces difficulties such as technical infeasibility, meaningless disclosure, user gaming and intellectual property right infringement. And personal data empowerment faces difficulties such as personal difficulty in exercising data rights and excessive personal data empowerment, making it difficult for big data and algorithms to operate effectively. Anti-algorithmic discrimination faces difficulties such as non-machine algorithmic discrimination, impossible status neutrality, and difficult realization of social equality. Taking scenarios of algorithms lightly is the root cause of the traditional algorithm regulation path dilemma. Algorithms may differ in attributes due to specific algorithmic subjects, objects and domains involved. Therefore, algorithm regulation should be developed and employed based on a case-by-case approach to the development of accountable algorithms. Following these development principles, specific rules can be enacted to regulate algorithm disclosure, data empowerment, and anti-algorithmic discrimination.