Knowledge discovery in databases can be enhanced b y zntroducang '%atalytic relations" conveying external knowledge. The new information catalyzes database inference, manifesting latent channels. Catalytic inference as zniprecase in nature, but the granularity of inference may be fine enough to create security conipromises . Cat a1 yt ic inference is coinpu t a t ion a1 1 y intensive. However, it can be automated by advanced search engines that gather and assemble knowledge from information repositories. The relentless inforination gathering potential of such search engines makes them forniidable security threats. This paper presents a formalism for modeling and analyzing catalytic inference in "mixed" databases containing vartous precise, zmprecise and fuzzy relations. The iriference forrnalisrri is flexible and robust, and well-suited to amplementation.