Cybercrime thrives on malicious websites and URLs, which host and propagate phishing, spear phishing, ransomware, and zero-day malware attacks. As modern attackers use cloud and DevOps tools like CI/CD pipelines, they can develop and deploy malicious websites blazing fast. Existing web classification approaches, which largely depend on known lists of malicious URLs and reputation intelligence, cannot detect and classify newly encountered malicious URLs due to their low prevalence. Sophos' latest deep learning approach to detecting never-before-seen malevolent web pages achieves a 97.5%* detection rate at a 0.1% false positive rate. This provides security vendors with the speed and accuracy they need to support a variety of deployments like endpoints, firewalls, and web proxies.