Fala com um especialista
PreçosNotíciasFala com um especialista
Voltar à sala de imprensa
Ameaças

Weekly Threat Roundup: VPN Exploitation, AEM RCE Chain, NVIDIA Container Escape and Stealers Hunting AI Keys

Ransomware crews are exploiting a PAN-OS VPN bypass, an AEM flaw chain reaches unauthenticated RCE, an NVIDIA driver bug enables container escape, and new infostealers target AI coding-assistant keys. Here's what matters and what to do.

4T4Tify Team4Tify
Oct 15, 2026·7 min de leitura
Weekly Threat Roundup: VPN Exploitation, AEM RCE Chain, NVIDIA Container Escape and Stealers Hunting AI Keys

Weekly Threat Roundup: VPN Exploitation, AEM RCE Chain, NVIDIA Container Escape and Stealers Hunting AI Keys

This week's security news has a common thread: attackers rarely need anything exotic. A VPN bug, a stolen employee login, a webhook pointed at the wrong address or a forgotten plugin folder is often enough. Below is our selection of the stories that matter most for security teams, followed by what we recommend doing about them.

Actively exploited: patch these first

PAN-OS GlobalProtect / Prisma Access (CVE-2026-0257). An authentication bypass disclosed in May is now being used by the Qilin and Settra ransomware operations. Because the intrusion comes in through the victim's own VPN, it can look like ordinary remote work until data theft or encryption starts. Researchers warn that patching alone does not evict an attacker who is already inside: active sessions should be terminated and unexpected connections investigated (including sessions from hostnames such as kali).

SiteGuard Server Edition WAF (CVE-2026-106931, CVSS 8.5). A command injection flaw affecting version 2.20 and later (with statistics enabled, which is the default) is being exploited in the wild and can yield root/SYSTEM command execution. Who is behind it and who is targeted is not yet known.

Critical and high-severity vulnerabilities

  • Adobe Experience Manager (cloud). Four chained flaws (unauthenticated DAM bulk import, stored XSS, unsafe XML parsing and Groovy execution via /bin/asynccommand) lead to unauthenticated RCE. Adobe has fixed them on the cloud service.
  • NVIDIA GPU Display Driver for Linux (CVE-2026-47595, "Bad UVM", CVSS 7.5). An unprivileged user can write to memory that should be read-only, which a researcher turned into local root escalation and container escape. This matters for any shared GPU host or Kubernetes node running GPU workloads.
  • Harbor registry (GHSA-2phj-cp9f-qq6w, CVSS 8.5). An SSRF lets any user who can create a project point a webhook at the cloud metadata endpoint and obtain the server's IAM credentials.
  • WebAlliance eCommerce (CVE-2026-108065, CVSS 9.8). A critical SQL injection exposes customer PII and passwords and can lead to OS command execution if the database account is highly privileged. No patch has been announced.
  • Samsung QA-Board (CVE-2026-110392, CVSS 9.8). A shell command injection allows unauthenticated RCE in this experiment-tracking tool. It is now patched.
  • NSD DNS server. Four high-severity flaws (CVE-2026-101337/101338/101339/101341) could allow arbitrary command execution. Fixed in NSD 4.15.3.
  • Xen hypervisor (CVE-2026-79610). A grant-table race condition could let a guest escalate to the host. All versions from 4.0 onward are affected; the fix is in XSA-517, with configuration mitigations available.
  • MyReport Updater 2.1.1 (Windows/Citrix). A standard user can redirect the updater to a malicious server and get files written as SYSTEM, leading to DLL loading and full control of the server.

Malware and threat activity

  • Infostealers now target AI developer tooling. Aoki goes after credential and session files for AI coding assistants (Claude, Codex, Cursor, Codeium, Gemini). Warden Stealer, a Rust-based malware-as-a-service, does the same and adds a crypto clipper. Stolen API keys and OAuth tokens let attackers act under a developer's identity, reach connected repos and CI/CD pipelines, and resell model access.
  • Self-healing WordPress backdoor. A malicious must-use plugin reinstalls itself after removal, hides a rogue admin account, harvests admin passwords and spreads to other WordPress sites on the same server. It locates its command servers through an Ethereum smart contract ("EtherHiding"), which makes takedowns harder.
  • Noir wallet drainer. This kit uses DNS TXT records (resolved over public DoH) to find its backend. Instead of asking for a seed phrase, it tricks users into granting spending permissions that look harmless.
  • RAT activity. XWorm has grown into a surveillance platform (GPS, live microphone, webcam, and browser-session hijacking that bypasses MFA). KGB RAT is sold with ransomware and "harassment" features such as jump scares and forced BSODs. CikaBika (Rust, Discord C2) and BotHelper (.NET, live screen capture, AMSI patching) are also active. CastleStealer variants now bypass Chromium's app-bound encryption.
  • Meta GitHub intrusion. A stolen employee credential gave unknown actors access to some of Meta's GitHub repositories. About 1,500 people, mostly US employees, are being notified.

Defensive moves worth noting

Google is expanding Android scam protections, including real-time verification of bank calls, and has raised Chrome bounty rewards for memory-safety bugs. Mozilla launched Firefox Enterprise. Microsoft is urging organizations to start inventorying PKI, HSM and certificate dependencies now to prepare for post-quantum authentication.

What to do

  1. Palo Alto customers: patch CVE-2026-0257 now, disable authentication-override cookies you don't need, kill all active GlobalProtect sessions after patching, and hunt for unusual VPN logins. Treat any confirmed exposure as a possible breach, not just a patching task.
  2. Run a patch sweep across SiteGuard, NSD (4.15.3 or later), Xen (XSA-517), NVIDIA Linux GPU drivers, Harbor and MyReport Updater. Prioritize anything internet-facing or multi-tenant.
  3. WebAlliance eCommerce: until a fix ships, put the application behind a WAF with SQLi rules, cut the database account down to least privilege, and monitor for unusual queries.
  4. Harbor and other cloud workloads: enforce IMDSv2 or block access to the metadata endpoint from application containers, and restrict who can create projects and webhooks.
  5. Protect AI developer credentials: keep API keys out of plaintext dotfiles, use short-lived and scoped tokens, rotate keys on any endpoint compromise, and alert on unusual usage or billing spikes.
  6. WordPress: review wp-content/mu-plugins/ and the administrator list on every site, and enforce file-integrity monitoring.
  7. Identity first: the Meta incident shows again that one stolen login can be enough. Enforce phishing-resistant MFA on code hosting and SSO, and review long-lived tokens.

Not sure where you stand? A 4Tify vulnerability assessment can show which of these issues affect your perimeter, and our analysts review every finding before it reaches you.

Reporting summarized from the weekly ThreatsDay Bulletin by The Hacker News (15 Oct 2026), with 4Tify analysis and recommendations.

PARTILHAR