CVSS 9.9 Triple Threat, AI-Powered Android Malware, and Anthropic Strikes Back — Feb 21 Threat Digest

February 21 is a heavy news day. Three separate CVSS 9.9 critical vulnerabilities are being actively exploited in the wild simultaneously — BeyondTrust Remote Support (CVE-2026-1731), Roundcube webmail (CVE-2025-49113), and the continued fallout from the Cline CLI 2.3.0 supply chain hit. BeyondTrust is the most alarming: Palo Alto Networks Unit 42 confirmed active exploitation across financial services, healthcare, legal, and higher education in the US, France, Germany, Australia, and Canada — attackers are deploying VShell web shells, establishing C2 infrastructure, and exfiltrating data. CISA separately added two Roundcube flaws to its KEV catalog today, with the CVSS 9.9 deserialization bug weaponized within 48 hours of disclosure. On the AI frontier, researchers at BleepingComputer flagged PromptSpy — the first known Android malware to use generative AI at runtime, leveraging Google's Gemini model to adapt its persistence behavior per device. Meanwhile Anthropic launched Claude Code Security in limited preview, an AI-powered codebase scanner that finds vulnerabilities and suggests patches — a direct counter-offensive to the same AI-driven attack automation threat actors are now deploying at scale.