Why don't we use a URL-based phishing detector?