CAS numbers need to be validated for the ~4000 chemical pages. Since the only authoritative source is the American Chemical Society, SciFinder looks like the best bet. For various reasons (see previous IRC discussions), it is not practical for one editor to validate them all.
X. Shuai, Z. Jiang, X. Liu, and J. Bollen. Proceedings of the 13th ACM/IEEE-CS Joint Conference on Digital Libraries, page 25--28. New York, NY, USA, ACM, (2013)