A content sanitization process performed via a transcoding engine. The transcoding engine can include one or more patterns, each pattern identifying malicious content in a well-formed document. The transcoding engine also can include an annotation processor configured to annotate portions of the well-formed document which match the patterns. Finally, the transcoding engine can include a content sanitization processor configured to normalize the annotated portions in a transcoded document. In a preferred aspect of the invention, the transcoding engine can include a privacy policy resolution processor configured to identify terms of a privacy policy and to annotate portions of the well-formed document which violate the terms.