ShareFile Shutdown Orders, Citrix Bleed 2 Ransomware, and AI Coding Assistants You Can't Trust

A security tool somewhere is finding bugs faster than any patch team can close them. That's the optimistic framing. The less optimistic version: attackers have identical tools, identical speed, and zero obligation to file a ticket before they act.

This week's roundup covers the predictable fallout. Trusted software turning hostile. Last year's CVEs still doing damage because remediation queues move slower than exploit kits. Fake installers. Poisoned packages. AI assistants cheerfully executing instructions they shouldn't have received in the first place.

Nothing exotic. That's what makes it exhausting.

Threat of the Week

Progress Software has told customers running ShareFile Storage Zone Controllers on Windows to pull the plug on those servers. The company pre-emptively disabled access to affected accounts while it investigates what it calls a credible external threat. No specifics on what that threat actually is, and Progress is insisting there's no confirmed unauthorised access to accounts or data. Given the wording, expect that story to develop.

Zimbra Patches a Stored XSS That Could Gut Your Mailbox

Zimbra pushed out an update for its Classic Web Client addressing a critical flaw where a specially crafted email could execute arbitrary scripts the moment a user opens it. Session data, mailbox contents, account settings, all potentially exposed. No CVE assigned yet. The fix is out, the risk is real, and the update queue is not your friend here.

Jscrambler's npm Package Was Weaponised

Attackers used a compromised npm publishing credential to slip malicious versions of the Jscrambler package into the wild. The payload was a Rust-based infostealer targeting Windows, macOS, and Linux simultaneously, pulling developer credentials and secrets. This overlaps with activity JFrog previously documented under the name IronWorm, which has apparently expanded scope considerably since its first appearance.

Microsoft Details GigaWiper, an Iranian-Linked Backdoor With Three Ways to Kill a Machine

Microsoft published analysis of a post-compromise backdoor called GigaWiper, linked to an Iran-nexus threat actor. It comes with three distinct destruction modes: full disk wipe, Windows drive overwrite, or fake ransomware that encrypts files using a key it discards immediately, making recovery impossible. It also screenshots, screen records, and opens covert VNC sessions. Arson with extra steps.

SHELLSTORM: 1.4 Million Domains, 27 CVEs, One WordPress Plugin Massacre

A large-scale operation attributed to a Chinese or Chinese-speaking threat actor exploited 27 known CVEs across WordPress plugins to plant web shells on compromised servers at scale. Over 1.4 million domains were targeted, with the heaviest concentration of infections in Taiwan, the US, Germany, France, and the UK. The shells were used to drop the SNOWLIGHT loader and the VShell backdoor. This is not subtle work.

HalluSquatting: When AI Hallucinations Become an Attack Vector

Researchers have documented a technique called HalluSquatting that turns AI coding assistants' tendency to invent package names into a reliable attack path. The method involves monitoring what package names AI agents hallucinate, registering those names for real, embedding malicious code in them, and waiting. Pair that with prompt injection and you have an AI dutifully fetching and running code the user never asked for. Dependency confusion always had legs. This gives it new ones.

Citrix Bleed 2 Is Being Used to Deploy DragonForce Ransomware

Huntress has documented half a dozen intrusions in the first half of 2026 where attackers exploited CVE-2025-5777, Citrix Bleed 2, across unrelated organisations using the same seven-step playbook. The pattern is consistent: gain access, escalate to SYSTEM via a registry symlink trick targeting the AppMgmt service, create rogue local admin accounts, establish persistence through legitimate remote tools like ScreenConnect and Zoho Assist, then deploy DragonForce ransomware. The standardised playbook suggests an organised operation rather than opportunistic one-offs. Patch exposed NetScaler appliances. Kill the outstanding sessions. Audit for accounts you don't recognise.

CVEs You Should Be Looking At This Week

The heavy hitters include U-Boot vulnerabilities BRLY-2026-037 through BRLY-2026-042, a cluster of Ubiquiti UniFi flaws, multiple BeyondTrust Remote Support and Privileged Remote Access issues, GhostLock (CVE-2026-43499), Januscape in the Linux KVM/x86 stack (CVE-2026-53359), a Microsoft Edge RCE, IBM WebSphere Application Server bugs, PHP issues, OWASP ModSecurity flaws, a Palo Alto PAN-OS vulnerability, Windows HTTP.sys exposure (CVE-2026-47291), Roundcube webmail, Foxit PDF Reader, GitLab CE and EE, and a problem in OpenAI Codex for macOS (CVE-2026-14898). Patch in urgency order. Start with anything internet-facing.

Fake VPN Installer Drops a Full-Featured RAT

An MSI file disguised as an installer for the Kuailian VPN (also known as LetsVPN) has been delivering GoodPersonRAT, a full remote access trojan with keylogging, browser manipulation, auto-update capability, and persistent access. Nothing surprising about fake installers delivering RATs. The interest here is the specific VPN brand being spoofed and the sophistication of the payload.

Fake Braintree NuGet Package Was Skimming Live Payment Data

A malicious .NET package named Braintree.Net impersonated the legitimate Braintree SDK and deployed an implant that intercepted live payment card data, stole Braintree merchant API keys, and harvested environment secrets on load. It included sandbox detection and only activated its payment skimming logic in production environments. Someone built this carefully.

RedHook Android Trojan Abuses Wireless ADB for Shell Access

An updated version of the RedHook Android trojan has expanded its capabilities to abuse Android's wireless ADB debugging feature to gain shell-level access without user interaction. It retains core RAT functions like screen streaming and keylogging while adding more aggressive privilege abuse. Originally focused on Vietnam, it's now targeting users in Indonesia. Distribution runs through spoofed government and financial sites, with the actual APK payloads hosted on AWS S3 and GitHub to boost delivery reliability and dodge detection.

Helix Extortion Group Using Vishing and MFA Abuse Against SharePoint

A newly emerged data extortion crew called Helix is reportedly spawned from the BlackFile and ShinyHunters ecosystems. Their approach splits across two compromised identities: one quietly exfiltrates data from SharePoint over days or weeks, while a separate account is compromised later solely to deliver the extortion demand via Microsoft Teams and email internally. The second account does no exfiltration. It exists purely to post the ransom note inside the victim's own environment. Operationally clean, unfortunately.

Microsoft: Expect More Security Updates Because AI Finds Bugs Faster Now

Microsoft has told customers to brace for an uptick in Windows security patches. The reason is that the company is using AI-assisted techniques to surface vulnerabilities faster than traditional methods allowed. The logic is sound: find it before attackers do. The practical reality is that the same acceleration in discovery applies to anyone else running similar tooling. More patches is probably good news. It would be better news if the gap between patch release and patch applied wasn't quite so wide.

Cryptominers Found in Compromised AI Gateways

Threat actors have been breaking into LiteLLM Proxy instances connected to Amazon Bedrock and using them to run cryptomining payloads. Initial access came through internet-exposed SSH. The broader observation from Darktrace is more important than the mining payload: AI infrastructure sitting at the intersection of cloud services, identity, and compute is a high-value target that most organisations are not treating as such. It should be subject to the same scrutiny as any other critical infrastructure component.

Rare Werewolf Targets Russian Aerospace Firms

A spear-phishing campaign attributed to Rare Werewolf (also known as Librarian Ghouls) is targeting aerospace organisations in Russia, using spoofed domains impersonating legitimate research institutes. Password-protected attachments ultimately configure AnyDesk for silent unattended access, exfiltrate the AnyDesk configuration to an attacker-controlled email account, and establish persistence. The group is known to operate across Russia, Belarus, and Kazakhstan.

Tools Worth Knowing About

Praetorian released Caeruleus, an open-source Go binary for Bluetooth Low Energy testing that consolidates scanning, GATT interaction, fuzzing, and security checks into a single tool with JSON output for scripting. Useful if you're doing BLE work and tired of stitching together half a dozen separate utilities.

PhantomFS is a Windows honeypot that uses the Projected File System to create convincing decoy files in memory, fake credentials, SSH keys, financial documents, none of which ever touch disk. When something opens one, it logs the event and fires a desktop alert with filename, process, and timestamp. High-confidence detection, no ML required. Read the code before deploying it anywhere that matters.

The Pattern This Week

Every efficiency we built to ship faster has become someone else's entry point. The package you didn't scrutinise. The remote access tool left running after the job finished. The AI agent that executes whatever it reads. We created the shortcuts. Other people are using them.

Patch the urgent items. Kill the sessions you forgot were active. Find out what's still exposed to the internet that has no business being there. Not exciting work. Just the work that matters before something else breaks.