💥 BREACHES & INCIDENTS
- UNM Health Sciences ransomware attack disrupts patient care; Byod gang claims breach of 36,000 systems
The Byod ransomware group attacked the University of New Mexico Health Sciences Center, forcing ambulance diversions and surgery cancellations before systems were restored.- Patients, staff, and students of UNM Health Sciences Center and UNM Health hospitals and clinics in New Mexico were affected, including the state's only Level I trauma center, UNM Hospital.
- The attack on October 26, 2026 forced ambulance diversions, suspension of outpatient surgeries and appointments, and cancellation of health sciences classes; UNM says only end-user devices were affected, not core systems or servers.
- The new Byod ransomware group claimed compromise of 36,000 systems, including Active Directory, Group Policy, the entire Microsoft Entra tenant, and two Oracle databases.
- Byod claims it stole student and employee data including Social Security numbers, HR and payroll records, internal emails, research data, and Cerner EHR rosters; it offered free decryption keys for hospital systems and up to $2,500 per affected patient while keeping the stolen data.
- UNM has not confirmed the claims, says it made no concessions to the threat actors, and warned the community not to engage with anyone claiming responsibility.
🕵️ RESEARCH & DEEP DIVES
Microsoft: North Korean threat actors are operationalizing AI across their tradecraft
Microsoft Threat Intelligence reports North Korean threat actors are adopting AI at high levels to accelerate and scale their operations.- Applies to organizations targeted by North Korean actors: cryptocurrency and financial firms, IT companies and software supply chains, government entities, and education-related organizations.
- AI is used across targeting and reconnaissance, social engineering and identity deception, malware and tooling development, and operational scaling.
- Jade Sleet, Coral Sleet, and Jasper Sleet have 'operationalized' AI across workflows; Diamond, Onyx, Pearl, and Sapphire Sleet are 'accelerating'; Emerald Sleet is 'expanding'.
- Recent examples: Jade Sleet ran campaigns through three front companies in a week, Onyx Sleet generated dozens of tailored lures in minutes, and Coral Sleet quickly built tooling to trojanize private source code repositories.
- Onyx Sleet and Pearl Sleet use the open-source agentic AI penetration testing framework ARTEX; Microsoft assesses AI-enabled operations remain human-directed, not autonomous.
Eclypsium warns attackers can tamper with AI models via firmware and infrastructure
Eclypsium warns that threat actors can sabotage AI models by tampering with the firmware and software of the infrastructure running them.- Applies to organizations deploying AI models on systems from Nvidia, AMD, Intel and Arm, including frontier AI labs.
- AI model runtime state is controlled by a hierarchy of software, firmware, memory controllers, management processors and I/O devices, each a tampering point.
- Host machines are vulnerable to ordinary malware such as firmware rootkits; third-party firmware update services, orchestration agents and management controllers create additional attack paths.
- Compromising the management plane or DPUs can enable firmware modification and direct control of attached workloads; a demo changed runtime model weights in memory while model files, hashes and signed artifacts remained valid, fooling integrity checks.
- Eclypsium partnered with Nvidia on the research and recommends DPUs be used to enforce security policies independently of the AI system.
Rust zip crate flaws (RUSTSEC-2026-0367, CVE-2026-115905) enable privilege escalation and RCE
CVE-2026-115995
Two vulnerabilities in the Rust zip crate — a setuid/setgid permission-bit flaw and a CVSS 10.0 zip-cloaking RCE — are fixed in zip 9.0.5.- Affects the widely used Rust zip crate; versions up to 9.0.0-pre3 are vulnerable to RUSTSEC-2026-0367 (GHSA-vv5m-jxxh-4627, CWE-732, CVSS 7.3), fixed in 9.0.0.
- The reader copies setuid, setgid, and sticky bits from archive entries instead of masking them, letting a crafted archive plant a setuid-root file when extracted by a privileged process (uploaders, installers, CI/CD, container builds).
- CVE-2026-115905 (GHSA-m665-v7pm-7grj, CVSS 10.0) is a 'zip cloaking' flaw: ZipArchive trusts the central directory while streaming readers trust the local header, enabling arbitrary file overwrite and RCE.
- Both bugs are fixed in zip 9.0.5, the version to pin.
- Cosmonic demonstrates containment of these flaws using a deny-by-default, capability-driven Wasm sandbox (wasmCloud), with a non-weaponized demo repo on GitHub.
📋 ADVISORIES
📄 Source for Yoba malware: Fake ChatGPT and Claude apps abuse signed executables to deploy XMRig miners — blog.checkpoint.com
📄 Source for Pre-auth RCE in TightVNC Viewer 2.8.88 via framebuffer integer overflow, fixed without a CVE — ssd-disclosure.com
📄 Source for Alpine Linux nearly hacked via GitLab CVE-2026-97229; GitLab rebuilt from scratch — docs.gitlab.com
🔓 CVEs & KEV
CVE-2026-115805 — Rapid7 Insight Agent — CVSS 7.8 — Rapid7 Insight Agent Local Privilege Escalation via Shell Injection in mcp.js...
CVE-2026-118763 — sigstore sigstore-js — CVSS 7.4 — @sigstore/oci sends the registry Authorization header to a cross-host blob-up...