The Deep Purple Sec – January 2026
A monthly round-up of interesting articles, reports and tools for tech experts, covering both offensive and defensive topics.
🔴 Red Team
📝 Achieving Remote Code Execution on n8n via Sandbox Escape – CVE-2026-1470 & CVE-2026-0863
JFrog Security has identified two critical n8n sandbox vulnerabilities: CVE-2026-1470 (9.9 Critical) allows arbitrary JavaScript execution via AST sanitisation flaws, enabling remote code execution (RCE). CVE-2026-0863 (8.5 High) affects Python execution in “Internal” mode, allowing OS command execution and full RCE.
The sandbox can be bypassed: CVE-2026-1470 exploits a flaw in the handling of JavaScript with and constructor to execute arbitrary code. CVE-2026-0863 exploits format-string introspection and Python 3.10+ AttributeError.obj to access restricted built-in functions and execute OS commands.
Fixes are available in the following n8n versions: CVE-2026-1470 fixed in 1.123.17, 2.4.5, 2.5.1; CVE-2026-0863 fixed in 1.123.14, 2.3.5, 2.4.2. Users should upgrade promptly to mitigate risks.
These findings emphasise the need for ongoing reassessment of sandboxes and robust defence-in-depth for untrusted code, highlighting the challenges involved in safely sandboxing dynamic languages such as JavaScript and Python.
📌 Source: https://research.jfrog.com/post/achieving-remote-code-execution-on-n8n-via-sandbox-escape
📝 Unveiling VoidLink – A Stealthy, Cloud-Native Linux Malware Framework
VoidLink is a cloud-native malware framework designed to provide long-term access to Linux systems, particularly in cloud and container environments. Written in Zig, it features a modular design with over 30 plugins for customised attacks, utilising OPSEC features such as runtime code encryption, self-deletion and adaptive behaviour, as well as user-mode and kernel rootkits.
It detects major cloud providers (AWS, GCP, Azure, Alibaba, Tencent) and container platforms (Kubernetes, Docker), harvesting cloud credentials and Git secrets, and targeting software engineers for espionage or supply-chain attacks. Its web dashboard manages implants and plugins, and monitors compromised systems, covering reconnaissance, credential theft, lateral movement, persistence and anti-forensics.
VoidLink highlights threats to Linux cloud infrastructure, emphasising the importance of proactive defence. Its stealth capabilities, rootkits and modularity pose a challenge to detection and response, as it uses eBPF and LKM rootkits to conceal processes and network activity, underscoring the need to monitor unusual activity, secure credentials and harden containers.
📌 Source: https://research.checkpoint.com/2026/voidlink-the-cloud-native-malware-framework
📝 Can AI identify zero-day vulnerabilities?
The Equixly blog details how their AI agent discovered CVE-2026-0773, a critical RCE vulnerability (CVSS 9.8) resulting from unsafe deserialisation via the cloudpickle library used in AI frameworks. It allows unauthenticated attackers to execute arbitrary code by exploiting Python’s insecure pickle deserialisation. The AI autonomously explored APIs, sequenced operations and generated context-aware inputs, identifying an unauthenticated endpoint deserialising pickle data – unlike traditional scanners, which fail to detect such issues due to a lack of context.
The post warns of the risks posed by unsafe deserialisation in AI frameworks and emphasises the importance of treating all serialised data as untrusted. It recommends safer formats (safetensors, JSON, Protocol Buffers) and cryptographic signing. The CVE highlights the role of AI-driven security testing in identifying critical vulnerabilities missed by conventional methods.
One example is CloudPickle’s serialisation of executable lambdas, which can be exploited to achieve remote code execution (RCE). This reflects a trend of pickle-based CVEs in AI/ML, highlighting the ongoing risks posed by unsafe deserialisation.
📌 Source: https://equixly.com/blog/2026/01/14/can-ai-identify-0days
📝 Task Failed Successfully – Microsoft’s “Immediate” Retirement of MDT
The SpecterOps blog has revealed critical vulnerabilities in Microsoft’s Deployment Toolkit (MDT), prompting its decommissioning rather than patching. The vulnerabilities include the ability to locate MDT servers and bypass security via XXE attacks in the unauthenticated MDT monitoring service, which uses System.Xml.XmlDocument in versions of the .NET Framework prior to 4.5.2.
Attackers can exploit the monitoring service’s API to register clients, inject XXE payloads, and trigger authentication coercion or leak sensitive files such as CustomSettings.ini, thereby compromising the credentials of the privileged MDT service account.
MDT’s retirement highlights the need to migrate to supported solutions, disable the monitoring service, segment servers, audit permissions and apply the principle of least privilege. Risks associated with unsupported software remain as vulnerabilities go unpatched.
🔵 Blue Team
📝 Synthetic Data: A New Frontier for Cyber Deception and Honeypots
The Resecurity blog discusses the use of synthetic data for cyber deception and honeypots. This fabricated data mimics real patterns without revealing any real information, helping to mislead attackers and gather threat intelligence safely.
Key points include honeypot accounts using synthetic datasets such as fake consumer records and payment transactions. Resecurity used over 28,000 fake records and 190,000 fake transactions to uncover attacker tactics that are effective even against advanced threat actors.
Synthetic data provides a controlled way to study attackers, gather intelligence and strengthen defences. It also enables decoy apps and accounts that mimic high-value targets to improve threat detection.
📝 THE KNOWNSEC LEAK: Yet Another Leak Revealing China’s Contractor-Driven Cyber-Espionage Ecosystem
The Knownsec leak in late 2025 revealed a Chinese cybersecurity firm linked to state espionage, offering tools such as GhostX (browser exploits, credential theft), Un-Mail (email surveillance) and Passive Radar (network mapping). Backed by the ZoomEye and TargetDB datasets, it targets critical infrastructure in Taiwan, Japan, India and the West.
The leak exposed Knownsec’s data lake containing breached global data (LinkedIn, Yahoo, Facebook, IDs, telecoms records) used for deanonymisation and social engineering. This enables technical assets to be linked to identities for rapid intrusions. For defenders, it highlights China’s sophisticated cyber operations and the need for robust network segmentation, monitoring of DNS and routing changes, and good credential management, particularly in sensitive sectors. The integrated toolset demonstrates how state actors utilise commercial firms to project cyber power, thereby complicating attribution and defence.
📝 A closer look at a BGP anomaly in Venezuela
Cloudflare’s analysis of the BGP route leak involving Venezuela’s CANTV (AS8048) shows that it was likely caused by poor routing export/import policies, rather than malicious intent. Since December, eleven leaks have affected multiple prefixes, with AS8048 acting as the source of the leak. Routes from AS6762 (Sparkle) were redistributed to AS52320 (V.tal GlobeNet), indicating a Type 1 hairpin leak.
AS prepending repeated AS8048 in the path makes routes less attractive, suggesting a misconfiguration rather than traffic interception. Route leaks are common in South America and are unrelated to Nicolás Maduro’s capture.
This incident highlights the need for robust routing policies and standards such as RFC 9234 and ASPA to prevent leaks and secure BGP by enforcing customer-provider and peer-peer roles.
📌 Source: https://blog.cloudflare.com/bgp-route-leak-venezuela
Never trust, always check