A dataset being as popular as their's is will contaminate the data just by people discussing it and creating their own public test sets of similar problems.

Still, probably not that much compared to employees targeting it.