February 21 is delivering a heavy threat cycle. BeyondTrust's critical CVE-2026-1731 (CVSS 9.9) is being actively exploited in the wild by threat actors deploying web shells, VShell backdoors, C2 infrastructure, and conducting lateral movement and data exfiltration. Palo Alto Unit 42 confirmed the campaign is hitting financial services, legal, healthcare, and higher education sectors across the US, France, Germany, Australia, and Canada. This is a full-spectrum intrusion playbook running on a single unpatched flaw.
On the supply chain front, Cline CLI 2.3.0 was poisoned via a compromised npm publish token on February 17, silently installing OpenClaw — a self-hosted autonomous AI agent — on developer machines via a postinstall hook. No malicious payload observed beyond unauthorized installation, but the attack vector is textbook and the implications for AI-assisted dev tooling are serious. Meanwhile, two Roundcube webmail flaws (CVE-2025-49113 at CVSS 9.9, CVE-2025-68461 at CVSS 7.2) were weaponized within 48 hours of public disclosure and are now on CISA's KEV catalog.
The defensive side got a boost: Anthropic launched Claude Code Security in limited research preview, positioning AI as a proactive vulnerability scanner for enterprise codebases. The move signals a shift — the same AI capabilities being weaponized by attackers are now being formally deployed as defenders. Also notable from the HackerNews feed: Anthropic's "Making frontier cybersecurity capabilities available to defenders" post is trending hard, suggesting the industry is watching this pivot closely.