Critical Gogs and n8n RCE flaws demand urgent patching

Critical remote code execution flaws in Gogs and n8n head this week's list of patched vulnerabilities, and both deserve immediate attention from anyone running self-hosted development or automation tools. Gogs, a Git service, has a maximum severity issue tracked as CVE-2026-52813 with a CVSS score of 10.0. Aikido Security discovered that organization names containing path traversal sequences such as ../ are accepted, so repositories under those names are written to arbitrary filesystem locations. An attacker can create a nested structure of Git repositories, overwrite another repository's hooks configuration, and achieve remote code execution, meaning they can run their own commands on the server. The fix is version 0.14.3, which also addresses CVE-2026-52810, a logic bug that allowed writes to read-only repositories, and GHSA-6vxv-wg6j-5qwp, a cross-site scripting flaw in the outdated jsvine/notebookjs library used to render Jupyter notebook files. In n8n, an open-source workflow automation platform, authenticated users with permission to create or modify workflows can exploit a prototype pollution vulnerability, a JavaScript flaw that lets an attacker change the shared template of objects, in the XML and GSuiteAdmin nodes. That leads to remote code execution on the n8n instance, and it has been assigned CVE-2026-33696 with a CVSS score of 9.4. The researcher Simon Koeck discovered the flaw, and n8n fixed it in versions 2.14.1, 2.13.3, and 1.123.27.

Signed driver abuse continues to give attackers a way past endpoint defenses without needing to bring their own vulnerable driver, a technique known as BYOVD. Check Point reverse engineered Microsoft Defender's Boot-Time Removal driver, BTR.sys, and showed it can be repurposed as a universal kernel operation engine. The key is a golden window between system start and user mode initialization, which lets a tool like BTR_CLI mimic the legitimate Windows Defender remediation process while bypassing endpoint security. Because BTR.sys carries a valid Microsoft signature, signature-based blocking alone does not stop it, researcher Jiří Vinopal noted. In a separate campaign, Grandoreiro malware is abusing the legitimate Duplicate Files Finder application to run malicious code through DLL sideloading, a technique that places a harmful library where a trusted app expects a legitimate one. Acronis telemetry shows the activity remains concentrated in Latin America, with Mexico, Spain, Peru, and Argentina accounting for most infections. The initial sample includes sandbox detection, virtual machine checks, process blacklisting, and environment profiling before it contacts command-and-control (C2) servers, which suggests avoiding analysis is a high priority for the operators. Another delivery chain combines ErrTraffic with Cruciferra. ErrTraffic, sold by a threat actor named LenAI, is a malware-as-a-service and traffic distribution system that pushes threats through compromised WordPress websites, ClickFix social engineering lures, and EtherHiding. WatchGuard says it has recently delivered Remus Stealer, Vidar Stealer, Okobot, LegionLoader, OnionDrop-related payloads, and BabaDedaLoader. eSentire observed victims landing on compromised WordPress sites injected with an obfuscated ErrTraffic JavaScript loader. That loader queries a Polygon smart contract to resolve its command-and-control domain and then requests the next stage, which serves a ClickFix lure. Cruciferra then uses a legitimate but vulnerable driver, DCRCVDrv.sys, in a BYOVD attack to escalate privileges and terminate security processes. The final goal is to launch Remus Stealer through process hollowing.

Industrial and building systems also appeared in this week's disclosures. Claroty's Team82 found 23 vulnerabilities in Copeland XWEB Pro refrigeration controllers, and some can be chained to bypass security mechanisms and achieve root-level remote code execution. A compromised controller could remotely manipulate cooling fans and compressors and conceal a temperature increase while food spoils. Danfoss AK-SM 800A refrigeration controllers had multiple issues too, including a hidden code-of-the-day authentication mechanism that could bypass normal authentication and a command-injection vulnerability leading to remote code execution. Another flaw let authenticated users inject arbitrary Nginx configuration directives, which could manipulate web traffic and cause a denial-of-service condition. Both vendors have fixed all identified issues. On a single corporate workstation, Gen Digital discovered a hand-written 12 KB Windows backdoor while hunting for unusual WMI persistence. The malware disguised itself as legitimate Realtek software and had a limited command set. Its most unusual feature was that the address of its command-and-control server was not stored as readable text or encrypted data but encoded in the number of trailing spaces on each line of a fake desktop.ini file. To a user, and to many automated inspection systems, the file would appear almost empty, but to the malware those spaces spelled out its server address. There is no evidence connecting the backdoor to a known threat actor, and the absence of related samples suggests it may have been a deliberately targeted operation.

AI-related privacy and abuse concerns rounded out the week. OpenAI announced a preview of Private Safety Processing for select customers, an approach that monitors model misuse without retaining customer data. For zero data retention deployments, customer content remains on infrastructure the customer controls, and OpenAI is also developing an option where content is stored on OpenAI infrastructure encrypted with keys controlled by the customer. In both cases automated systems can identify potential misuse and return limited safety signals without exposing prompts or responses to OpenAI personnel. Google separately showcased Homomorphic Encryption Intermediate Representation, or HEIR, an open-source compiler toolchain for homomorphic encryption, a method that lets models work on encrypted inputs without decrypting them. On the unregulated side, a service called Kriminal AI offers paying subscribers answers without filters or guardrails, claiming more than 2,300 users and subscriptions from $12.99 to $99.00 per month. It is openly accessible on the clearnet with a no filters, no guardrails tagline. ThreatDown says the service appears to use Grok for primary inference, Google Cloud and Cloudflare for hosting, Anthropic Claude for a long-context model layer, Llama routed through OpenRouter for some specialized tasks, Tavily for live search, NowPayments for cryptocurrency checkout without know-your-customer, and Cloudflare and Let's Encrypt for DNS and TLS. Apple also agreed to change its App Tracking Transparency consent prompts in Germany after the Federal Cartel Office found the feature gave Apple's own apps more favorable prompts than third-party apps. The changes will apply in almost a