February 21 is shaping up as one of the heaviest exploit cycles of the year. CISA added two Roundcube webmail vulnerabilities to its KEV catalog — CVE-2025-49113 (CVSS 9.9, remote code execution via deserialization) and CVE-2025-68461 (CVSS 7.2, XSS via SVG animate tag). The 9.9-rated flaw was diffed and weaponized within 48 hours of public disclosure, per Dubai-based FearsOff. That turnaround time is a textbook OPSEC window indicator — attackers are running near-automated patch-diff pipelines now.
BeyondTrust CVE-2026-1731 (also CVSS 9.9) is being actively exploited in the wild. Palo Alto's Unit 42 documented the full kill chain: network recon → web shell → VShell/C2 → backdoor installs → lateral movement → data theft. Targets span financial services, legal, healthcare, higher education, and tech across the US, France, Germany, Australia, and Canada. This is a wide-net campaign, not targeted espionage. On the supply chain front, Cline CLI 2.3.0 pushed OpenClaw to developer machines via a compromised npm publish token — another reminder that dev tooling is prime real estate for threat actors.
The one bright spot: Anthropic launched Claude Code Security in limited preview, an AI-powered codebase scanner that finds vulnerabilities and suggests patches. The timing is deliberate — Anthropic explicitly framed it as a counter to adversaries using AI to automate vuln discovery. HackerNews is also flagging the US Supreme Court striking down Trump's global tariffs as the political story of the day (1381 points), and ggml.ai joining Hugging Face as a major open-source AI consolidation move. The defender-attacker AI arms race is now fully out in the open.