This Week in Cybersecurity: Sandworm Targets Cisco Firewalls, a CVSS 10 E-Commerce Zero-Day, and AI That Hacks Credentials in Six Hours

Explore this content with AI:

ChatGPT | Perplexity | Claude | Google AI Mode

Russian Sandworm and Qilin ransomware are actively exploiting critical Cisco firewall flaws

Cisco's Secure Firewall Management Center (FMC) is under active attack from two distinct threat clusters, according to disclosures from Cisco Talos and Help Net Security reporting published September 10. At the center of the activity are two vulnerabilities: CVE-2026-20079, an authentication bypass flaw that allows remote, unauthenticated attackers to execute commands with root access via specially crafted HTTP requests, and CVE-2026-20316, a static-credentials vulnerability that lets unauthenticated attackers remotely log in through a hardcoded low-privilege account. CVE-2026-20079 was disclosed and patched in early March 2026; CVE-2026-20316 was disclosed on July 29. Neither patch has been applied universally.

Talos identified three active intrusion clusters. The first — currently unattributed — exploits CVE-2026-20079 to deploy malicious web shells and JAR files for credential harvesting. The second is Sandworm, the Russian military intelligence (GRU) unit, which is leveraging either vulnerability to establish reverse shells, harvest firewall configurations, and deploy implants that enable persistent command execution and credential theft. The third is Qilin, a ransomware operation that uses the static credentials in CVE-2026-20316 for network reconnaissance, credential theft, and ransomware staging. Organizations running unpatched FMC deployments are effectively handing attackers the keys to their perimeter security architecture.

Why it matters: Sandworm targeting your firewall management plane is categorically different from a network intrusion — attackers with FMC access can read your entire network topology, pivot to managed devices, and disable detection. If your organization runs Cisco FMC, treat CVE-2026-20079 and CVE-2026-20316 as incident-response priorities, not standard patch-cycle items. Verify patching status now and audit FMC logs for web shell activity and anomalous JAR file creation.

Read more at SecurityWeek

StyleSmuggler: a CVSS 10.0 Adobe Commerce zero-day is actively dropping a Rust backdoor

E-commerce security firm Sansec published research on September 5, naming a new zero-day: StyleSmuggler, tracked as CVE-2026-75650 (CVSS 10.0, APSB26-146). The vulnerability enables unauthenticated remote code execution on Adobe Commerce and Magento 2.4.9 systems — no credentials or user interaction required. Sansec confirmed attacks began in September 2026, with multiple malicious payload variants observed in the wild, including a Rust-compiled backdoor and a PHP web shell dropped on compromised storefronts. Adobe shipped an emergency patch (APSB26-146), and CISA added CVE-2026-75650 to its Known Exploited Vulnerabilities catalog this week.

The Rust backdoor component is particularly notable: Rust-compiled malware is increasingly favored by advanced threat actors for its performance characteristics and the difficulty it presents to memory forensics tools. Attackers with an active StyleSmuggler compromise have full command execution and persistent access to the underlying server — and to any payment, customer, and transaction data flowing through it. Merchants running unpatched Magento 2.4.9 instances should treat this as immediate remediation, not a scheduled maintenance window.

Why it matters: A CVSS 10.0 score with confirmed in-the-wild exploitation and a Rust backdoor payload is as serious as it gets for e-commerce operators. If your organization runs Adobe Commerce or Magento — directly or through a vendor — verify your version and patch status today. Treat CISA's September 18 remediation deadline for federal agencies as the maximum timeline, not the target.

Read more at The Hacker News

Microsoft's record September Patch Tuesday includes two exploited Windows zero-days and a wormable DNS flaw

Microsoft's September 2026 Patch Tuesday addressed 974 vulnerabilities — a record single-month total — including two actively exploited Windows zero-days and a DNS flaw security researchers are calling the spiritual successor to SigRed. The two exploited zero-days are CVE-2026-81963 (CVSS 7.8), a Windows Update Stack link-following vulnerability that allows authenticated local attackers to escalate to SYSTEM, and CVE-2026-85880 (CVSS 7.8), a Windows Advanced Local Procedure Call (ALPC) heap buffer overflow enabling the same SYSTEM-level privilege escalation via AppContainer escape — affecting Windows 10 and older Windows Server versions (2012 through 2022). Both are added to CISA KEV this week.

The wormable DNS flaw (CVE-2026-69730) drew particular attention from researchers as a "SigRed successor": it is among 20 newly patched vulnerabilities classified as wormable, meaning a remote, unauthenticated attacker could achieve arbitrary code execution with no user interaction. Also in this release: CVE-2026-69676, a Kerberos authentication bypass that enables remote code execution on domain controllers, and CVE-2026-55007, an Exchange Server RCE triggered by malicious Visio attachment processing. The 974-CVE total includes 723 Windows flaws and 222 Office bugs, and Microsoft noted that 20 of the newly resolved vulnerabilities could be "weaponized to propagate without user involvement."

Why it matters: Two actively exploited LPE zero-days mean attackers already have working code — any user-level foothold in a Windows environment can now be elevated to SYSTEM with tools that are already circulating. Patch CVE-2026-81963 and CVE-2026-85880 first. Then address the wormable DNS flaw CVE-2026-69730 and the Kerberos DC exploit CVE-2026-69676 before the end of the week. The record patch count underscores that attack-surface growth is outpacing remediation capacity for most organizations — prioritization frameworks aren't optional anymore.

Read more at SecurityWeek

Google GTIG: an AI coding assistant built an autonomous credential-harvesting operation in under six hours

A Google Threat Intelligence Group (GTIG) report published this week documented a financially motivated threat actor who built a fully autonomous credential-harvesting framework using an AI coding chatbot — and deployed it against healthcare, government, and media cloud environments, compromising thousands of API credentials in less than six hours. The threat actor used preconfigured markdown instruction sets as operational playbooks, with the AI system autonomously managing vulnerability scanning, real-time troubleshooting, and IP rotation. No ongoing human direction was required once the initial prompt and instructions were set.

The attack specifically targeted cloud API credentials and co-opted victim environments for unauthorized AI workloads — a pattern that has been growing since late 2025, when compromised cloud accounts are used to run large-model inference at victims' expense. GTIG's characterization is direct: the framework "operated without handholding by a human," demonstrating a speed advantage that fundamentally changes attacker economics. Where a manual credential-harvesting campaign might take days and require sustained human effort, this framework executed the equivalent operation in hours. Sectors targeted — healthcare, government, media — align with organizations that hold high-value credentials for sensitive cloud environments.

Why it matters: The threat here isn't AI "hacking" in science-fiction terms. The threat is automation at scale: attackers can now compress multi-day reconnaissance and exploitation operations into a single autonomous execution cycle. Security teams operating on a 24-hour detection-and-response cadence are structurally behind. Audit your cloud API credential exposure, rotate high-value keys on a regular cycle, and review your detection coverage for anomalous API calls and unusual compute usage — the latter being a leading indicator of credential co-option for AI workloads.

Read more at The Hacker News

JetBrains' own cloud service was breached through an unpatched TeamCity flaw — and backup secrets were taken

JetBrains disclosed on September 3 that its Cadence cloud service was compromised between August 8 and August 24, 2026, via CVE-2026-63077, a CVSS 9.8 unauthenticated remote code execution vulnerability in TeamCity On-Premises versions prior to 2025.11.7 and 2026.1.3. JetBrains had not applied the patch to its own infrastructure. Attackers extracted AWS IAM credentials, project source code, configuration files, and critically, secrets from a 2024 server backup — including cloud credentials, source control tokens, package and container registry credentials, and deployment keys. JetBrains' investigation, which concluded September 3, could not fully determine the scope of unauthorized access to customer S3 storage.

The incident carries a specific irony: TeamCity is a CI/CD platform used by development teams to build, test, and deploy software, and JetBrains — its creator — was breached through an unpatched version of its own product. Threat activity was described as "consistent with both criminal and advanced persistent threat (APT) activity," though no specific attribution has been made. For organizations using TeamCity, the primary concern is whether the exfiltrated deployment keys and registry credentials were used to access downstream build pipelines or artifact repositories.

Why it matters: When a CI/CD vendor's own infrastructure is compromised through their own unpatched product, it surfaces two problems at once: the vendor's patching discipline and the scope of secrets that live in build systems. Review your TeamCity deployment for CVE-2026-63077 patch status immediately. Separately, audit secrets stored in any build configuration JetBrains' Cadence service had access to — rotate deployment keys, registry credentials, and cloud tokens associated with any project that touched Cadence. Treat the backup secrets exposure as a potential supply chain entry point until ruled out.

Read more at The Record

Final thoughts

This week's stories share a throughline that's becoming uncomfortably familiar: attackers are targeting the infrastructure organizations use to protect and build their own infrastructure. Cisco's FMC isn't a target of opportunity — it's the thing that manages the firewalls. JetBrains' Cadence isn't a random cloud service — it holds the keys to development pipelines. StyleSmuggler doesn't just compromise a website; it compromises the payment and customer data flowing through it. Attackers know the highest-leverage entry points aren't end-user devices anymore — they're the security appliances, build systems, and platforms the rest of the organization depends on.

The Google GTIG report on autonomous AI agents adds a dimension that changes the tempo of all of this. The window between a vulnerability's disclosure and its exploitation at scale used to be measured in days or weeks. When a single automated session can survey, exploit, and harvest thousands of credentials in under six hours, detection latency and response times matter more than ever.

Check back next Thursday for the latest in cybersecurity news and analysis.