How Advanced URL Filtering Works

Learn how URL categorization and inline machine learning work on firewalls with URL filtering enabled.
Where can I use this?
What do I need?
  • Prisma Access
  • PAN-OS
  • Advanced URL Filtering license
    For Prisma Access, this is usually included with your Prisma Access license.
Advanced URL Filtering classifies websites based on site content, features, and safety. A URL can have up to four URL categories, including risk categories, that indicate the likelihood that the site will expose you to threats. As PAN-DB, the Advanced URL Filtering URL database, categorizes sites, firewalls with Advanced URL Filtering enabled can leverage that knowledge to enforce your organization’s security policies. In addition to the protection offered by the PAN-DB database, Advanced URL Filtering provides real-time analysis using machine learning to defend against new and unknown threats. This provides protection against malicious URLs that are updated or introduced before URL filtering databases have an opportunity to analyze and add the content, giving attackers an open period from which they can launch precision attack campaigns. Advanced URL Filtering compensates for the coverage gaps inherent in database solutions by providing real time URL analysis on a per request basis. The ML-based models used by advanced URL filtering have been trained, and are continuously updated, to detect various malicious URLs, phishing web pages, and C2.
Websites that indicate the presence of certain advanced threats are additionally processed though a cloud-based inline deep learning system, using detectors and analyzers that complement the ML-models used by Advanced URL Filtering. Deep learning detectors can process larger data sets and can better identify complex malicious patterns and behaviors through multi-layered neural networks. When Advanced URL Filtering receives HTTP response data from the firewall upon receipt of a suspicious web request, the data is further analyzed through the deep learning detectors and provides inline protection against evasive zero-day web attacks. This includes cloaked websites, in which web page contents are surreptitiously retrieved from unknown websites—this can include malicious content that URL databases are unable to account for, multi-step attacks, CAPTCHA challenges, and previously unseen one-time-use URLs. Because evasive malicious websites are in a constant state of flux, the detectors and analyzers used to categorize websites are updated and deployed automatically as Palo Alto Networks threat researchers improve the detection logic, all without requiring the administrator to download update packages.
When a user requests a web page, the firewall queries user-added exceptions and PAN-DB for the site’s risk category. PAN-DB uses URL information from Unit 42, WildFire, passive DNS, Palo Alto Networks telemetry data, data from the Cyber Threat Alliance, and applies various analyzers to determine the category. If the URL displays risky or malicious characteristics, the web payload data is also submitted to Advanced URL Filtering in the cloud for real-time analysis and generates additional analysis data. The resulting risk category is then retrieved by the firewall and is used to enforce the web-access rules based on your policy configuration. Additionally, the firewall caches site categorization information for new entries to enable fast retrieval for subsequent requests, while it removes URLs that users have not accessed recently so that it accurately reflects the traffic in your network. Additionally, checks built into PAN-DB cloud queries ensure that the firewall receives the latest URL categorization information. If you do not have Internet connectivity or an active URL filtering license, no queries are made to PAN-DB.
The firewall determines a website’s URL category by comparing it to entries in 1) custom URL categories, 2) external dynamic lists (EDLs), and 3) predefined URL categories, in order of precedence.
Firewalls configured to analyze URLs in real-time using machine learning on the dataplane provides an additional layer of security against phishing websites and JavaScript exploits. The ML models used by local inline categorization identifies currently unknown and future variants of URL-based threats that match the characteristics that Palo Alto Networks has identified as malicious. To keep up with the latest changes in the threat landscape, local inline categorization ML models are added or updated via content releases.
When the firewall checks PAN-DB for a URL, it also looks for critical updates, such as URLs that previously qualified as benign but are now malicious.
If you believe PAN-DB has incorrectly categorized a site, you can submit a URL category change request in your browser through Test A Site or directly from the firewall logs.
Did you know?
Technically, the firewall caches URLs on both the management plane and the dataplane:
  • PAN-OS 9.0 and later releases do not download PAN-DB seed databases. Instead, upon activation of the URL filtering license, the firewall populates the cache as URL queries are made.
  • The management plane holds more URLs and communicates directly with PAN-DB. When the firewall cannot find a URL’s category in the cache and performs a lookup in PAN-DB, it caches the retrieved category information in the management plane. The management plane passes that information along to the dataplane, which also caches it and uses it to enforce policy.
  • The dataplane holds fewer URLs and receives information from the management plane. After the firewall checks URL category exception lists (custom URL categories and external dynamic lists) for a URL, the next place it looks is the dataplane. Only if the firewall cannot find the URL categorized in the dataplane does it check the management plane and, if the category information is not there, PAN-DB.

Recommended For You