A method and system are provided for filtering harmful HTML content from an electronic document. An application program interface (API) examines the fundamental structure of the HTML content in the document. The HTML content in the electronic document is parsed into HTML elements and attributes by a tokenizer and compared to a content library by a filter in the API. The filter removes unknown HTML content as well as known content that is listed as harmful in the content library. After the harmful HTML content has removed, a new document is encoded which includes the remaining safe HTML content for viewing in a web browser.