This paper identifies potential security loopholes in the implementation of support for meta-predicates. Closing these loopholes depends on three conditions: a clear distinction between closures and goals, support for an extended meta-predicate directive that allows the specification of closures, and the availability of the call/2-N family of built-in meta-predicates. These conditions provide the basis for a set of simple safety rules that allows meta-predicates to be securely supported. These safety rules are currently implemented by Logtalk, an object-oriented logic programming language, and may also be applied in the context of Prolog predicate-based module systems. Experimental results illustrate how these rules can prevent several security problems, including accidental or malicious changes to the original meta-predicate arguments and bypassing of predicate scope rules and predicate scope directives.