Algorithms and protocols are the product of research, and are published to invite the scrutiny of peer review. Just as a high school algebra student may find himself unable to identify his own mistake in his algebra, developers may be subject to blind spots in their evaluation of their own methods.
When these methods are tested by other minds, obvious failings become immediately evident, and more subtle errors are observed by those who compromise the method using some innovative attack, or test.
The most common misconception in security, is to presume that introducing proprietary modifications, improves the implementation through added “secrecy.” The argument: “No one can guess my innovation by random speculation.”
While this is true, the secrecy of the modification leads to false confidence.
Compare the discussion to the security of duplicated ID. A passport is very hard to duplicate, analogous to a proprietary implementation. A credit hard is less hard to duplicate, but can be de-certified by the single phone call of anyone claiming that the number has been published. “I found this card, lying in the street, and I just want to make sure the number isn’t misused.” Somewhere in the middle, is a state Driver’s License, that can be re-issued to change the address.
For an identity thief, a stolen passport is very difficult to counterfeit (or duplicate,) but as long as no one knows he has done so, the effort exerted to falsify it is rewarded proportionally, by the credibility invested in the document (due to the difficulty of counterfeiting it, and the seriousness of the associated penalties.)
Fake ID’s are more common in bars, and the penalties associated with counterfeiting them are lower, so Driver’s Licenses are not as trusted as passports, while remaining very worthwhile to certify identity, but also to counterfeit.
A proprietary implementation has a recursive flaw. The modification is not widely scrutinized, so it may be subject to the above mentioned mental blind spot, and since mistakes are relatively commonplace, few authorities are willing to associate greater penalties, with the compromise of such a modification.
Every algorithm has to be implemented is some way, and the act of “implementing” introduces details that cannot be standardized. The experience and meticulousness of the person or team preparing the implementation will affect the sea-worthiness of the final product.
Impartial review, by equally qualified critics, is a good practice, with the understanding that such rules as Kerckhoff’s Principle will safeguard the user from associated compromise.