Security news.
Autonomous AI tools have crossed into active offensive operations, with threat actors deploying agents to discover and exploit zero-day vulnerabilities in medical infrastructure and orchestrate data theft at cloud providers. Meanwhile, federal law enforcement and enterprise identity systems are reeling from high-profile breaches, and CISA has expanded its Known Exploited Vulnerabilities catalog with actively targeted enterprise flaws. Cloud and infrastructure security teams must also contend with critical vulnerabilities in core Linux hypervisors and widely used libraries like Ghostscript and GnuPG.
AI Agents Discover and Exploit Zero-Day in Medical Imaging Software
Suspected China-linked attackers used autonomous Cairn AI agents to locate a zero-day in open-source MONAI Label versions 0.8.5 and earlier, leveraging path traversal to gain unauthorized root server access.
FBI Investigating Hack of National Criminal Justice Database
Federal officials are probing an intrusion into the FBI's National Crime Information Center (NCIC), a core system used across local, state, and federal agencies to exchange sensitive law enforcement data.
CISA Adds Actively Exploited VMware, Cisco, and SAP Flaws to KEV Catalog
CISA issued mandates covering active exploitation in VMware vCenter, Cisco Expressway series hardcoded credentials (CVE-2026-76466), and SAP NetWeaver (CVE-2025-42884).
Rackspace Spot Cloud Breach Exposes Customer Kubernetes Secrets
Attackers used compromised credentials to penetrate Rackspace's Spot cloud platform, allowing them to spin up unauthorized compute workloads and read Kubernetes secrets within tenant clusters.
Microsoft Entra Connect Health Flaw Exploited to Exfiltrate Tenant Configs
Threat actors exploited a vulnerability in Microsoft Entra Connect Health to breach tenant data, compromising administrator email addresses and internal identity configuration details.
OpenRouter Removes AI Providers Over Undetected Months-Long Prompt Theft
Security researchers uncovered months of stealth prompt interception affecting three AI inference providers, capturing sensitive user inputs even when Zero Data Retention (ZDR) protections were supposedly active.
Ghostscript Sandbox-Escape Flaw Enables Server RCE via Crafted Uploads
A vulnerability in the widely deployed Ghostscript file-processing interpreter allows malicious document uploads to escape internal sandboxes and execute arbitrary commands on host systems.
Linux KVM Hypervisor Bug Exposes Multi-Tenant Memory on AMD Hosts
Tracked as CVE-2026-99312, a kernel KVM flaw allows malicious guest virtual machines with nested virtualization enabled to dump gigabytes of host memory, risking cross-tenant data exposure on shared cloud infrastructure.