Security news.
Active exploitation targeting developer tooling and AI infrastructure tops today's updates, alongside high-severity flaws hitting refrigeration controllers and IRC servers. CISA and honeypot operators are reporting immediate threat activity against recently disclosed vulnerabilities, while attackers increasingly weaponize autonomous AI offensive frameworks. Security teams should prioritize patching exposed endpoints and reviewing access tokens.
Active Honeypot Scans Target LiteLLM and Starlette RCE Chain
Honeypots are seeing requests targeting CVE-2026-42271 and CVE-2026-48710, which can be chained together to achieve unauthenticated remote code execution. Security teams running exposed LiteLLM deployments should apply mitigations immediately.
CISA Adds Actively Exploited UnrealIRCd Flaw to KEV Catalog
Attackers are actively weaponizing CVE-2026-114713, a high-severity command injection vulnerability that allows execution of arbitrary commands with server or IRCOp privileges. CISA has formally mandated remediation for federal systems.
Unpatched RCE Vulnerabilities Threaten Danfoss Refrigeration Controllers
Six zero-day flaws in Danfoss AK-SM800A controllers permit complete remote device takeover without authentication, with firmware 5.4.0 remaining vulnerable. European cybersecurity authorities have issued advisories urging operators to isolate affected hardware from the internet.
Hackers Probe Disclosed Flaws in Copeland XWEB EVO Controllers
Attackers have begun targeting 14 newly disclosed vulnerabilities in Copeland refrigeration controllers shortly after public release. The targeting follows multi-agency government warnings regarding cyber threats against cold chain and industrial cooling infrastructure.
Decade-Old 'NoisyTCP' Linux Flaw Exposes Nginx and SSH to Crashes
An 11-year-old Linux kernel vulnerability dubbed NoisyTCP affects default configurations across most major distributions. The flaw allows remote attackers to trigger denial-of-service conditions on critical services like Apache, Nginx, and SSH.
Stolen AI Provider Key Compromises Cursor IDE Marketplace Extension
An access token stolen from infrastructure provider Fireworks AI was used to publish a malicious version of Anthropic's Claude Code extension on the Cursor marketplace. Developers using the affected extension should rotate compromised API keys and audit their environments.
FBI Warns of Global Attacks Powered by ARTEX AI Framework
Threat actors are increasingly leveraging the open-source ARTEX offensive AI tool to automate intrusion chains and scan for exposed secrets. The framework has recently been tied to supply chain compromises and lateral movement campaigns targeting enterprise cloud environments.
PostgreSQL Resolves 52 Flaws Including 22-Year-Old Code Execution Bug
PostgreSQL has patched CVE-2026-6460 (CVSS 8.8), an authenticated type-confusion flaw allowing users with object-creation rights to execute OS-level code. Full remediation requires updating the database server and recompiling any installed C-language extensions.