Daily Cybersecurity News – September 6, 2026
Unpatched Magento Zero-Day Backdoors Stores
CriticalWhat happened
Attackers began exploiting an unpatched zero-day in Magento Open Source and Adobe Commerce on September 4, 2026, allowing unauthenticated remote code execution on store servers and installation of persistent backdoors.
Dutch firm Sansec discovered the flaw, named it StyleSmuggler, and published an early advisory on September 5 because stores were being compromised in real time. The chain was reproduced on clean installs of versions 2.4.7, 2.4.8, and 2.4.9. As of September 6 Adobe had issued no advisory, CVE, patch, or workaround.
Independent confirmation came from Disrex Group, which handled two fully compromised Magento Open Source stores and one failed attack; one victim ran a fully patched 2.4.6 line and another ran 2.4.8.
Who is affected
All current Magento Open Source and Adobe Commerce versions appear affected, including 2.4.9 and recently patched 2.4.6-p15 / 2.4.6-2026-aug lines. Sansec confirmed the unauthenticated chain on 2.4.7-2.4.9 Open Source installs.
Online stores of any size that expose GraphQL are at immediate risk. Classic and Hyvä storefronts can often disable GraphQL; headless and PWA setups generally cannot. Exact victim count is unknown but active exploitation is underway.
Why it matters
Successful exploitation yields full server code execution and a persistent backdoor, enabling theft of customer data, payment details, order records, and further malware deployment across the e-commerce estate.
Because no official patch or CVE exists yet and Adobe's next scheduled release is September 8, operators face a window of active mass exploitation with only temporary mitigations available. Compromised stores can become launch points for supply-chain or customer-facing attacks.
How it could have been prevented
Temporarily disable GraphQL if the storefront does not require it (most classic and Hyvä setups do not; headless/PWA usually do).
Monitor for anomalous admin or server activity, deploy web-application firewall rules that block the observed attack patterns, keep Sansec Shield or equivalent runtime protection enabled if licensed, and apply Adobe's forthcoming fix immediately once released. Rotate any credentials that could have been exposed on compromised hosts.
Relevant professional terms
- Zero-day
- A software vulnerability that is unknown to the vendor and has no official patch available when attackers begin using it.
- Unauthenticated RCE
- Remote code execution that an attacker can trigger without any valid login or session, giving immediate control of the target process or server.
JetBrains Cadence Breached via TeamCity Zero-Day
CriticalWhat happened
Unidentified threat actors exploited the critical TeamCity vulnerability CVE-2026-63077 (CVSS 9.8) to breach JetBrains' own Cadence cloud environments in August 2026. The flaw allows unauthenticated remote code execution via the agent polling protocol through deserialization of untrusted data.
JetBrains discovered the intrusion on August 23. Attackers accessed a 2024 Cadence server backup containing credentials and configuration, extracted multiple AWS IAM users and secrets, reached S3 storage with user data, and may have obtained source code synced from PyCharm projects. CISA added the CVE to its Known Exploited Vulnerabilities catalog on August 5.
Cadence is JetBrains' hosted service for running ML and heavy workloads on cloud GPUs from the PyCharm IDE.
Who is affected
Cadence users whose projects, credentials, secrets, email addresses, source code, or execution inputs/outputs resided on the affected servers. JetBrains stated the impact matches the group it previously contacted directly and found no additional users.
Also exposed: AWS IAM credentials belonging to JetBrains employees who used the service, plus files in related S3 buckets. Anyone still running unpatched TeamCity (before 2026.1.3 / 2025.11.7) remains at risk of the same RCE.
Why it matters
A vendor's own cloud CI/CD-adjacent service was compromised via a known critical RCE that was already under active exploitation, leading to credential and potential source-code theft. Stolen AWS keys and secrets can enable further cloud resource abuse or lateral movement.
Users must treat all prior Cadence executions and stored secrets as untrusted. The incident underscores the danger of unpatched build-server software even inside vendor environments.
How it could have been prevented
Immediately revoke and rotate every credential and secret ever used with Cadence executions or stored in the service. Treat all historical execution inputs, outputs, and project data as potentially compromised.
Patch TeamCity to 2026.1.3, 2025.11.7 or later without delay. Audit AWS IAM users and keys associated with Cadence, enable least-privilege policies, and monitor for anomalous cloud activity originating from those identities.
Relevant professional terms
- Credential rotation
- The practice of replacing passwords, API keys, tokens, or certificates with new ones so that any stolen copies become useless.
- Deserialization RCE
- A vulnerability in which untrusted serialized data is converted back into objects in a way that lets an attacker run arbitrary code during the reconstruction process.
Critical VMware Workstation Flaw Executes Host Code
HighWhat happened
Broadcom released patches for two flaws in VMware Workstation and Fusion. The more severe is an integer-overflow issue that lets a local attacker who already holds administrative privileges inside a guest VM execute arbitrary code on the host when the VM uses a VMXNET3 virtual network adapter.
A second stack-based buffer overflow in the HGFS component allows similar host-side code execution as the VMX process from a privileged guest. Both require prior admin rights inside the VM; those rights can be obtained via separate phishing or misconfiguration. No evidence of in-the-wild exploitation of these two bugs has been reported.
Affected releases are Workstation and Fusion 25H2 and 26H1; fixes ship in 26H1u1 builds. Broadcom states there are no workarounds.
Who is affected
Users and organizations running VMware Workstation or VMware Fusion versions 25H2 or 26H1, especially those that grant administrative rights inside guest VMs or use the VMXNET3 adapter and HGFS shared-folders feature.
Desktop and laptop developers, testers, and IT staff who run nested or privileged VMs are the primary population. Enterprise vSphere/ESXi hosts are not the subject of these particular advisories.
Why it matters
Guest-to-host escapes break the fundamental isolation guarantee of desktop virtualization. An attacker who compromises one VM can reach the underlying workstation, steal host data, install persistent malware, or move laterally.
VMware products have repeatedly been high-value targets; recent vCenter flaws saw rapid exploitation. Even though these two issues need prior guest admin rights, that precondition is commonly met after phishing or local privilege escalation.
How it could have been prevented
Upgrade immediately to VMware Workstation 26H1u1 and VMware Fusion 26H1u1. There are no workarounds.
Limit administrative privileges inside guest VMs, disable unneeded features such as HGFS shared folders when possible, keep host and guest fully patched, and treat any guest that processes untrusted content as potentially hostile to the host.
Relevant professional terms
- Guest-to-host escape
- A vulnerability that lets malicious code running inside a virtual machine break out and run code or access resources on the physical host machine.
- Integer overflow
- A condition where an arithmetic operation produces a value too large for the allocated storage, wrapping around and often leading to memory corruption that attackers can abuse.
5400 Hacked Sites Deliver Blockchain-Stored ClickFix
HighWhat happened
A large campaign has compromised more than 5,400 small-business websites, predominantly WordPress and PrestaShop, and injected scripts that pull next-stage payloads from smart contracts on the BNB Smart Chain Testnet. The technique is known as EtherHiding.
The injected code presents a ClickFix social-engineering lure: a fake CAPTCHA that instructs the visitor to open the Windows Run dialog and paste a PowerShell command, which then downloads and runs the final payload. Later variants replaced the ClickFix stage with a WebRTC data-channel stager that opens a covert encrypted channel to attacker infrastructure and executes received JavaScript entirely in browser memory.
Because the malicious configuration and code live on-chain, operators can update the payload at any time and the infrastructure is hard to take down.
Who is affected
Visitors to any of the 5,400-plus compromised small-business sites, especially Windows users who follow the ClickFix instructions. Site owners whose WordPress or PrestaShop installations were backdoored.
The initial compromise vector for the sites themselves remains unreported; the scale indicates opportunistic mass scanning or exploitation of common CMS weaknesses.
Why it matters
ClickFix turns ordinary users into the delivery mechanism by abusing legitimate OS features (Run dialog + PowerShell). Storing payloads on a public blockchain removes traditional C2 takedown options and lets attackers swap malware on demand.
The shift to in-memory WebRTC stagers further reduces disk artifacts, complicating detection. Compromised legitimate sites provide high-trust distribution that bypasses many reputation filters.
How it could have been prevented
For site owners: keep WordPress, PrestaShop, plugins and themes fully updated; remove unused extensions; deploy file-integrity monitoring and a web application firewall; scan for unexpected JavaScript includes or external script loads.
For end users and enterprises: train staff never to paste commands from web pages into Run or PowerShell; block or alert on suspicious PowerShell invocation; use browser isolation or hardened browsing; monitor for WebRTC anomalies and unexpected outbound peer connections. Network controls that restrict access to known malicious BSC endpoints can help.
Relevant professional terms
- ClickFix
- A social-engineering tactic that tricks a user into manually copying and executing a malicious command, often disguised as a CAPTCHA or fix instruction.
- EtherHiding
- The practice of storing malware payloads or configuration data inside blockchain smart contracts so the malicious content is resilient, publicly readable, and easily updated by the attacker.
REVSTEALER Modules Kill Defender and Mine Crypto
HighWhat happened
Elastic Security Labs documented four previously unreported persistent modules linked to the emerging Windows information stealer REVSTEALER. After the core stealer exfiltrates data and deletes itself, these modules remain: ProManager, WinUpdate, SoftManager, and LockAppHost.
They share packer, runtime resolution, and Polygon smart-contract configuration tradecraft with REVSTEALER. LockAppHost elevates via CMSTP, disables Windows Update services and tasks, adds broad Microsoft Defender exclusions, then runs a hidden cryptocurrency miner. The defensive weakenings persist even after the miner is removed.
ProManager targets desktop crypto wallets (including Electron-based ones) with overlay and keylogging behavior; WinUpdate performs clipboard crypto-address swapping and seed-phrase collection; SoftManager turns the host into a reverse proxy.
Who is affected
Windows endpoints infected with REVSTEALER or the associated activity set. REVSTEALER has been sold as a commercial stealer since at least February 2026 and targets browser credentials, crypto wallets, gaming accounts, messaging data, and files.
Anyone running desktop cryptocurrency wallets or whose machine can be turned into a proxy or miner is particularly impacted. Exact infection volume is not quantified in the report.
Why it matters
The stealer itself is designed to be non-persistent, yet it leaves behind a suite of durable implants that continue theft, fraud, proxying, and resource abuse while actively dismantling OS defenses. Disabling updates and Defender leaves the machine open to any follow-on malware.
Clipboard swap and wallet overlay techniques directly monetize crypto users. Shared blockchain configuration channels give the operators resilient command infrastructure.
How it could have been prevented
Deploy and tune endpoint detection that catches CMSTP abuse, unexpected Defender exclusion changes, disabled update services/tasks, and crypto-miner process injection. Keep Microsoft Defender and Windows Update operational and monitored for tampering.
Use application allow-listing, restrict administrative rights, monitor clipboard and wallet-process behavior, and hunt for the named binaries and their persistence locations (Run keys, scheduled tasks, logon scripts, services). Rotate any crypto or account credentials on systems that show signs of infection.
Relevant professional terms
- Infostealer
- Malware whose primary job is to harvest passwords, cookies, wallet files, and other sensitive data from an infected computer and send them to the attacker.
- Defense evasion
- Techniques used by malware to disable, bypass, or hide from security tools such as antivirus, endpoint detection, and automatic update services so it can keep running.
Numbat: Open-Source AI Agent Observability Tool
LowHow it works
Numbat instruments AI agents to capture decision traces, tool calls, and runtime behavior.
- Collects prompts, intermediate steps, and final actions
- Surfaces unexpected or high-risk tool use
- Supports post-incident review and continuous monitoring of agent fleets
What happened
SANS Internet Storm Center highlighted Numbat, an open-source observability tool built for monitoring and understanding the behavior of AI agents. The diary entry introduces the project as a practical aid for operators who need visibility into autonomous or semi-autonomous agent actions, tool calls, and decision traces.
Details in the available crawl are limited to the ISC framing; the tool focuses on instrumentation and logging that help defenders and developers see what agents are doing in real time or after the fact.
Who is affected
Security teams, AI platform operators, and developers who deploy or defend LLM-powered agents, especially in environments where agents can call tools, access data, or take actions with real-world impact.
Organizations experimenting with agentic workflows that currently lack dedicated telemetry will find the most immediate relevance.
Why it matters
AI agents introduce new attack surfaces and failure modes that traditional application logging often misses. Without purpose-built observability it is difficult to detect prompt injection success, unexpected tool use, data exfiltration, or runaway behavior.
An open-source option lowers the barrier for teams that need to add agent-specific tracing and auditing before incidents occur.
Relevant professional terms
- Observability
- The ability to understand the internal state and behavior of a system by examining its external outputs such as logs, metrics, and traces.
- AI agent telemetry
- Structured recording of an autonomous agent's prompts, reasoning steps, tool invocations, inputs, outputs, and side effects so operators can audit and debug its actions.