The Week in One Line
An actively exploited, unauthenticated CVSS 10.0 Magento/Adobe Commerce RCE forced Adobe to issue an emergency hotfix; attackers exploited critical Langflow and Ruby on Rails flaws to gain code execution and harvest secrets; N-able released its fourth emergency hotfix in five weeks for a maximum-severity pre-authentication RCE in N-central; and CISA added actively exploited vulnerabilities across AI gateways, workflow orchestration, artifact management, HTTP frameworks, remote access, and communications infrastructure to its KEV catalog.
Key Takeaways
- E-commerce became the week’s highest-priority web-app incident. Adobe Commerce and Magento Open Source CVE-2026-75650, dubbed StyleSmuggler, is an actively exploited, unauthenticated RCE rated CVSS 10.0.
- Patch-only closure is not enough for compromised stores. Reporting on StyleSmuggler described PHP web shells and persistent Linux/Rust backdoors, so post-patch compromise assessment and credential rotation are essential.
- AI workflow platforms are high-value initial-access surfaces. Langflow CVE-2026-0768 was actively exploited for unauthenticated Python execution and reconnaissance against AI-provider, cloud, and Langflow secrets.
- Framework flaws can become broad secret-exposure events. Ruby on Rails CVE-2026-66066 can expose Rails secrets, cloud keys, database credentials, and API tokens, enabling broader compromise.
- Managed-service infrastructure remains a high-blast-radius target. N-able N-central CVE-2026-86218 affects builds before 2026.3.1.14; on-premises deployments need Hotfix 4 immediately.
- CISA KEV’s September 2 batch is directly relevant to AppSec: LiteLLM, Kestra, Starlette, JFrog Artifactory, SonicWall SMA1000, and Sangoma Switchvox all appeared in the actively exploited list.
Critical CVEs and Incidents
Deep Dives: The Flaws That Matter
Adobe Commerce and Magento: StyleSmuggler RCE (CVE-2026-75650, CVSS 10.0)
Adobe released Security Bulletin APSB26-146 for CVE-2026-75650, a zero-day affecting Adobe Commerce, Adobe Commerce B2B, and Magento Open Source. The issue can allow unauthenticated arbitrary code execution on an affected installation without user interaction.
Sansec named the issue StyleSmuggler and reported active exploitation against live stores from September 4. Reporting described deployment of PHP web shells and persistent Linux/Rust-based backdoors. Patching closes the exploit path, but it does not remove malware or persistence that may already have been installed.
Affected supported branches include multiple Adobe Commerce, Adobe Commerce B2B, and Magento Open Source 2.4.x releases. Verify the compatible VULN-39341 package and exact installation procedure through Adobe’s APSB26-146 guidance before deployment.
Action: Apply APSB26-146/VULN-39341 immediately. Hunt for unexpected PHP files, altered application code/templates, unknown Magento administrators, suspicious cron or systemd jobs, anomalous checkout/GraphQL/API traffic, and unusual outbound connections. Rotate application, database, payment-provider, cloud, and integration credentials.
Langflow: actively exploited AI-workflow RCE (CVE-2026-0768, CVSS 9.8)
Attackers actively exploited CVE-2026-0768 in Langflow, a low-code platform used to build AI applications and workflows. Insufficient validation of attacker-controlled input enables unauthenticated remote execution of arbitrary Python code.
Observed activity included reconnaissance for Langflow administration data and high-value environment variables such as OPENAI_API*, AWS_ACCESS*, AWS_SECRET**, Langflow secret material, SSH directories, and command history. A compromise can therefore become an AI-provider, cloud, or enterprise-API incident rather than a single-server event.
Action: Patch immediately, remove unnecessary public exposure, rotate all secrets reachable from the runtime, and monitor for sensitive-file enumeration, unexpected workflow changes, suspicious subprocesses, credential use, and abnormal egress.
Ruby on Rails: Previously patched flaw enters active-exploitation reporting (CVE-2026-66066, CVSS 9.5)
Although Rails disclosed and patched CVE-2026-66066 on July 29, it returned to this week’s threat picture after VulnCheck observed a parallel exploitation wave alongside Langflow activity, with canary hits in Singapore, Israel, and the UK. The flaw, dubbed KindaRails2Shell, arises from a discrepancy between Active Storage and the
libvipsimage-processing library that lets crafted uploads trigger arbitrary file reads.
Attackers can recover
secret_key_baseRails master keys, database passwords, cloud-storage credentials, and API tokens—enabling session forgery, account takeover, API abuse, database compromise, cloud access, and potentially follow-on RCE. VulnCheck identified more than 7,100 exposed vulnerable Rails instances in early August; the activity should be viewed as opportunistic credential harvesting across exposed application infrastructure, not necessarily evidence that the Langflow and Rails campaigns use the same operators.
Action: Upgrade Rails/Active Storage per vendor guidance. Rotate
secret_key_baseRails credentials, database passwords, cloud-storage keys, OAuth secrets, API keys, signing keys, and accessible cloud tokens. Review image-upload and Active Storage activity for suspicious file reads, credential exposure, and follow-on access. N-able N-central: maximum-severity RMM RCE (CVE-2026-86218, CVSS 10.0)
N-able released N-central 2026.3 Hotfix 4, build 2026.3.1.14, to address CVE-2026-86218, a critical pre-authentication remote-code-execution vulnerability in its remote monitoring and management platform. All builds before 2026.3.1.14 are affected.
Because N-central functions as an administrative control plane, compromise can create broad downstream exposure across managed endpoints and customer environments. N-able-hosted environments were patched by the provider; on-premises deployments require customer action.
Action: Upgrade on-premises N-central to 2026.3.1.14 / Hotfix 4. Review for unknown users, privilege changes, unexpected integrations, malicious scripts or automation, suspicious remote sessions, and anomalous managed-endpoint activity.
CISA KEV: AI, workflow, and application infrastructure under attack
CISA added seven actively exploited vulnerabilities to KEV on September 2. The group included LiteLLM MCP authentication weakness, Kestra unauthenticated command injection, Starlette request smuggling, JFrog Artifactory improper authentication, SonicWall SMA1000 SSRF/command injection, and Sangoma Switchvox SQL injection.
The concentration of AI gateways, workflow orchestration, developer tooling, remote-access systems, and HTTP frameworks is a clear indicator that application control planes and integration layers are now prime targets.
Action: Identify external LiteLLM/MCP, Kestra, Artifactory, Starlette, SonicWall, and Switchvox exposure. Patch first, then rotate connected tokens, audit new administrators/workflows/repository changes, and monitor for abnormal HTTP framing and suspicious API sessions.
Public PoC Watch: Telerik UI for ASP.NET AJAX
Progress patched Telerik UI for ASP.NET AJAX flaws in July, but public RCE exploit tooling was released on September 7. The chain targets certain non-default RadAsyncUpload configurations and combines CVE-2026-13182 with related flaws to enable unauthenticated code execution. No confirmed in-the-wild exploitation has been reported.vmtech+1
Action: Upgrade to Telerik UI for ASP.NET AJAX 2026.2.708 or later. If patching is delayed, disable the AsyncUpload handler if unused and monitor for suspicious IIS child processes, new .aspx files, or unexpected DLLs.
WordPress and CMS/API Security Watch
This week’s WordPress findings reinforce common API-security failures: REST routes with insufficient authorization, missing capability/nonce checks, unsafe upload handling, and server-side calls that expose or abuse third-party API credentials.
What to Do This Week
- Patch emergency web assets first: Adobe Commerce/Magento APSB26-146, N-able N-central Hotfix 4, Langflow, Rails/Active Storage, and all relevant CISA KEV assets.
- Treat StyleSmuggler as an incident-response event, not only a vulnerability-management task: patch, investigate for web shells/backdoors, rotate credentials, validate file integrity, and inspect payment, checkout, administrative, GraphQL, and API activity.
- Secure AI workflows and MCP systems: remove unnecessary public access, upgrade LiteLLM/Kestra/Langflow, scope and rotate model-provider/cloud/API credentials, and monitor tool execution and workflow changes.
- Prioritize high-blast-radius control planes: N-central, Artifactory, remote-access appliances, and workflow orchestration systems often hold administrative paths or credentials to many downstream systems.
- Harden reverse proxy and API paths: reject ambiguous HTTP framing; standardize proxy/origin parsing; validate schemas; enforce authentication, authorization, rate limits, and behavioral controls.
- Reduce CMS exposure: remove unused plugins, patch affected components, restrict public REST routes, block server-side script execution in upload locations, and monitor plugin/theme changes.
How Unified WAAP Protection Helps
This week’s incidents share a common theme: the initial foothold is commonly an exposed web endpoint, API route, storefront, management interface, workflow engine, developer service, or remote-access control plane. A unified WAAP program can reduce the exposure window and centralize response across web and API attack paths.
- Virtual patching for emerging RCE, SSRF, SQL injection, request smuggling, deserialization, GraphQL, and file-upload patterns.
- Continuous API discovery and governance to identify forgotten administrative APIs, undocumented REST routes, external MCP endpoints, and data-rich workflow interfaces.
- Behavioral bot and abuse detection for exploit reconnaissance, credential stuffing, malicious API automation, scraping, and business-logic abuse.
- Integrated L3/L4/L7 DDoS mitigation and centralized telemetry for blended attacks, anomaly investigation, and faster incident response.
Frequently Asked Questions (FAQ)
1. What is the most urgent web application issue this week?
CVE-2026-75650 (StyleSmuggler) in Adobe Commerce and Magento Open Source is the leading urgency. It is an actively exploited CVSS 10.0 unauthenticated RCE, with reports of PHP web shells and persistent backdoors. Patch immediately and conduct a compromise assessment.
2. Why should AI workflow platforms be treated as critical infrastructure?
Langflow, LiteLLM, and Kestra can hold model-provider credentials, cloud keys, enterprise API tokens, workflow definitions, data connections, and tool permissions. Their compromise can become an identity, data, cloud, or automation incident.
3. Is N-able N-central relevant to web application security?
Yes. N-central is a network-accessible administrative control plane. A pre-authentication RCE can create a high-blast-radius route to managed endpoints and customer environments.
4. What is virtual patching?
Virtual patching uses WAF/WAAP enforcement to block known exploit behavior while a vendor patch is tested and deployed. It reduces risk but does not replace remediation or post-compromise investigation.
5. How should we prioritize this week’s list?
Start with internet-facing and actively exploited systems: Magento/Adobe Commerce, Langflow, Rails applications, N-central, SonicWall SMA1000, Artifactory, Kestra, LiteLLM/MCP, Switchvox, and exposed Starlette applications. Then prioritize by business criticality, privilege, sensitive-data access, and connected credentials.
