ubisoft

This article is based on joint research with Eran Segal, researcher at Kodem Security. The most capable commercial AI models are now useful enough to attackers that they have become an integral part of their kill chain, in multiple steps. The Cybench benchmark tests models on offensive cyber tasks. Its current top performers (Claude Opus 4.6, Claude Sonnet 4.5, Grok 4) can write functional exploit code, reason through credential chains, and sustain complex reconnaissance workflows: multi-step offensive work that previously required human expertise. Malware families are already using this. Instead of generating a payload offline and shipping it, they wire a live LLM API into the malware itself so it can adapt its behavior at runtime on the infected host. Commercial providers run abuse detection and terminate accounts linked to malicious activity. A payment method creates a paper trail that investigators can follow. So attackers solve the access problem the same way they solve any resource problem: they steal it, find it free, or find it unguarded. This post covers five routes threat actors use to reach LLM inference without paying for it: buying offensive models on underground forums, using front-end models using 3rd party LLM service that allows paying in bitcoin, using free-tier or keyless public APIs, hunting for leaked API keys in developer artifacts, and exploiting self-hosted LLM servers left open on the internet. Method 1: Offensive LLMs and Anonymous Payment Cyber-oriented LLMs sold on underground forums are the most visible route. WormGPT, GhostGPT, KawaiiGPT, and Xanthorox are the most cited examples, covered in depth by Unit 42. These are returned open-weight models or jailbroken wrappers over commercial APIs, marketed specifically as having no content filter. They solve the moderation problem but not the cost problem: access is sold on a subscription basis, and the capability ceiling sits well below that of frontier commercial models. So they are useful for generating phishing content or simple malware stubs, but less so for the kind of autonomous multi-step offensive work that the frontier models in the Cybench ranking are capable of. Method 2: Using Frontier Models Through a Third-Party Service If a threat actor would like to use the frontier models to achieve top performance, they can still use these models using 3rd party services such as PayWithMoon and AIMLAPI. These services sit between the attacker and a commercial LLM provider, accepting cryptocurrency without identity verification and then funding a legitimate provider account on the attacker's behalf. The account itself reaches frontier models, but the funding trail stops at the middleman. The account will still get burned once abuse detection triggers, but replacing it is cheap. The upstream provider has no usable identity to pursue. This is how attackers buy frontier-model access while skipping the paper trail a normal commercial account would leave behind. Method 3: Free-Tier and Keyless Public Inference APIs A cheaper alternative to underground subscriptions exists in plain sight. Most major inference providers publish permanent free tiers that require nothing more than a disposable email address, and a handful of services accept requests with no credentials at all. An attacker who registers for a pool of free-tier accounts gets as meny tokens as he wishes without paying a dime. The scale of the free-tier ecosystem is easy to measure because the community has cataloged it. Public curated lists such as cheahjs/free-llm-api-resources and mnfst/awesome-free-llm-apis explicitly filter for providers that offer a permanent (not trial-credit) free tier with no credit card. Representative entries, with numbers pulled from each provider's own rate-limit documentation: Groq: 30 requests/minute (RPM) on all free-tier models, with requests/day (RPD) caps ranging from 1,000 (for the 70B llama) to 14,400 (for the 8B llama). Cerebras: 30 RPM, 14,400 RPD, and roughly 1M tokens/day on three of the four free-tier models (gpt-oss-120b, llama3.1-8b, qwen-3-235b). Cohere: 20 RPM on the Chat API and a hard cap of 1,000 total API calls/month on a trial key. Mistral La Plateforme: 1B tokens/month on the Experiment plan. No credit card is required, but a verified phone number is required, which is the highest sign-up friction in this group. HuggingFace: Free accounts are rate-limited on both the Hub API and the Inference API per 5-minute window. Anonymous per-IP access exists but is stricter than the free-account path. OpenRouter: 50 free model RPD with no deposit at all, and 1,000 RPD after a one-time $10 top-up that is never spent against model usage. SambaNova: 20 RPM and 20 RPD, with a 200,000 tokens/day cap. The tightest daily request ceiling in this group by a wide margin. These providers differ in rate limits, models, and throughput. What they share is that a usable credential requires nothing more than a disposable email address (a phone number in Mistral's case) and no payment method. Credentials can simply be rotated when limits are reached. The fully keyless end of the spectrum is thinner, but it exists. Pollinations.ai exposes an OpenAI-compatible endpoint that accepts requests with no authentication for basic use. DuckDuckGo's Duck.ai anonymizes browser-based access to Claude 3.5 Haiku, Llama 4 Scout, Mistral Small 3, GPT-5 mini, and GPT-4o mini with no account at all. These services are not designed for bulk programmatic use, but they are reachable from any HTTP client, and the only cost is rate-limit friction. Among the malware families in the intro table, LameHug/PROMPTSTEAL is the cleanest example of this route in the wild: it calls HuggingFace's Inference API for Qwen 2.5-Coder-32B-Instruct to drive reconnaissance and data theft, with no embedded credentials reported by Splunk. Whether the malware carries a token or registers one at runtime is not established, but either way, the enabling property is HuggingFace's no-credit-card free tier. Method 4: Exposed API Keys The fourth route to free model access doesn't require finding an exposed server at all. Developers routinely hardcode credentials directly into apps, config files, and scripts. These credentials can be found in GitHub in open-source projects, while closed-source projects contain the credentials in the app itself. These artifacts are submitted to VirusTotal when apps are submitted for malware analysis. It can be an APK, ELF, EXE, or any type of artifact shipped with the product. To find them systematically, we wrote a YARA rule targeting the key formats of the major AI providers: Google Gemini (AIzaSy…), OpenAI (sk-…), Anthropic (sk-ant-…), HuggingFace (hf_…), Replicate, Mistral, Cohere, Groq, and several others. We ran the rule as a retrohunting query across the VirusTotal corpus, collected the matching sample hashes, then pulled the raw files and ran a regex extraction pass to extract every key-value pair, provider, and surrounding code context. From there, we enriched each sample with VirusTotal metadata to understand detection rates and file types. The final step was validation: a lightweight GET request against each provider's model-list or whoami endpoint. No prompts sent, just a check of whether the key authenticates. The corpus yielded 647 unique keys across all providers. Roughly 62% were Google Gemini (AIzaSy…) keys. That concentration traces back to the Android developer ecosystem, where apps built for translation, search, or chatbot features commonly bundle the key directly in compiled resources or Java code. HuggingFace keys made up about 11%, Replicate about 8%, OpenAI sk- keys about 7%, and the remaining share was split across Voyage (5%), Mistral (3%), and Cohere (3%), with trace amounts of Anthropic, Groq, and OpenAI environment-style keys. The Mistral and Cohere keys concentrated heavily in a single file: a cracked "Collins Italian Dictionary MOD" Android APK that bundled 20 Mistral keys and 15 Cohere keys alongside 2 Gemini keys, with the small remainder scattered across two versions of a Ubisoft game APK. About 65% of the 659 unique samples are confirmed Android by VirusTotal's type classification. Another 18% are ZIP archives that follow the same submission pattern but were not explicitly tagged as Android. The true APK share sits between 65% and 84%. The remainder consisted of Windows PE files (5%), HTML pages, Python scripts, plain-text credential dumps, and a handful of Mach-O and ELF binaries. That Android skew isn't surprising. APKs are frequently submitted to VirusTotal for modding and repackaging, and their keys remain intact after decompilation. We submitted research samples to Intezer Analyze for code-based attribution, and three entries stand out. Four samples whose filenames suggested Akira ransomware are three Mimikatz binaries (1, 2, 3) and one malicious binary without family attribution, all credential-dumping tools that happened to carry API keys. The sample with a HuggingFace key is SolarMarker, an SEO-poisoning backdoor with infostealer capability. A Windows binary named SystemSettings.exe contained OpenAI, Replicate, and Voyage keys; the multi-key combination is more consistent with theft from a developer's machine than with intentional hardcoding. When we ran the validation, almost all the keys were dead. The revocation rate was approximately 99.5%, consistent with a corpus skewed toward older samples that had been on VirusTotal long enough to be detected, rotated, or simply expired. The small fraction that remained live consisted entirely of Google Gemini keys from Android APKs. All appeared to be genuine developer mistakes rather than exfiltrated credentials: a key embedded in a const in bundled JavaScript, one in a logging module in a compiled Android class, and one in a utility app's APK resources. Those three keys have been reported to Google. The method also illustrates why embedding API keys in client software is a particularly bad idea. Extracting a key from an APK requires a decompiler, and APKs have a reliable path to VirusTotal: users submit them for malware checks, repackaged versions circulate through third-party stores, and cracked builds get flagged automatically. The near-total revocation rate strongly suggests that LLM providers scan VirusTotal for their own key formats and automatically revoke matches. The three keys that were still live were all recent submissions, not yet caught by that sweep. If that pipeline exists, embedding a key in client-side code is not just a security mistake, but a futile one, and the key will likely be dead before it can be abused at scale. The takeaway for an attacker is that hunting VirusTotal for hardcoded keys is low-effort but low-yield. The more durable access method is the exposed LLM server. A server running vLLM (a popular open-source LLM inference framework) or an open Ollama instance requires no authentication, doesn't rotate anything while in use, and the owner usually doesn't know it's happening. Method 5: Hack Public LLM Hosting Servers Self-hosted LLM platforms make it easy to run your own models on your infrastructure, and that same ease extends to anyone who can access the port. Most ships have no authentication by default and expose administrative endpoints that let a stranger list installed models, queue inference jobs, load new models from remote URLs, or, in several cases, execute code on the host. When the server is exposed to the public internet, the attacker does not need a stolen key or a forum subscription. The victim is paying the GPU bill, carrying the API-key spend, or hosting the RCE. We scanned roughly 4,500 hosts across eleven of them. Every service had open instances, and 14 LocalAI hosts showed active compromise based on attacker-loaded model names consistent with a single automated campaign. The sections below cover what each platform is, how exposure gets abused, and what the scan found in the wild. Ollama Ollama runs open-weight LLMs locally. By default, it binds to 127.0.0.1, and the authentication is disabled. But setting OLLAMA_HOST=0.0.0.0 is a common step when accessing it from another machine on the network or from a frontend app running in a separate container. It exposes all interfaces, and anyone reaching it’s port gets full API, model management, and hardware access. SentinelOne Labs and Censys already published the definitive survey, documenting 175,000+ hosts chained into anonymous AI networks for free text, embedding, and bulk content generation on victim hardware. That pattern is now commercialized by Operation Bizarre Bazaar, which sells subscription access to a unified LLM gateway fronted by stolen Ollama endpoints, turning ad-hoc LLMjacking into a growing concern. LocalAI LocalAI is an OpenAI API-compatible model server supporting LLMs, image generation, speech, and transcription. Authentication is disabled by default. It also supports remote model installation, P2P distributed serving, and a built-in agent platform with support for MCP. Of all the services in this research, it has the widest attack surface. Of all the hosts scanned, 55% were confirmed open, the highest absolute count in this group. About 24% are API proxies with live upstream keys for OpenAI, Anthropic, and Google accessible to anyone who can reach the host. The most striking finding is evidence of automated exploitation at scale. About 21% of confirmed hosts carry model names with a consistent signature tied to ProjectDiscovery's nuclei scanner templates, with per-run timestamps mapping to late March and early April 2026. The pattern is consistent with automated scanning for an unauthenticated remote code execution path, in which a malicious URL supplied during model installation triggers server-side code execution. The exploit payload appears to load a small publicly available Italian-language model as a "hello world" confirmation, which recurs on every affected host. The markers not being cleaned up argue against mature attacker tradecraft. Operators running LocalAI can open /v1/models on their own host: any nuclei-rce-* or rce_ identifier is not human-chosen and indicates this campaign hit them. Langflow Langflow is a visual builder for multi-agent AI pipelines, widely used to prototype RAG systems and chatbots. Flows routinely embed hardcoded credentials: OpenAI and Anthropic API keys, database connection strings, Slack tokens, and webhook secrets. Anyone who can reach the host and read a flow config has all of them. Unlike the previous examples, this app does not have a known major misconfiguration, but it does not prevent attackers from being able to hack and gain access to this service. For example, two unauthenticated RCE bugs make reaching the config trivial: CVE-2025-3248:on the CISA KEV list, reliably patched only in 1.6.4+ CVE-2026-33017: fixed in 1.9.0, exploited in the wild within 20 hours of disclosure. Every confirmed host in our scan ran a version vulnerable to CVE-2026-33017; about 72% were also vulnerable to CVE-2025-3248. Several hosts didn’t authenticate at all, with flows, credentials, and both RCE paths openly accessible. Code execution on the Langflow host is the small prize. The keys inside the flows pivot to everything the workflows connect to. n8n n8n is a low-code workflow automation platform with 400+ service connectors and code execution nodes (workflow steps that run arbitrary scripts). It has the strongest default auth posture of any service in this research: User Management is enforced on fresh installs. But it does not prevent attackers from actively gaining access to n8n. Vulnerabilities such as CVE-2026-21858 ("Ni8mare", CVSS 10.0, fixed in 1.121.0), which is a vulnerability in the web hooks request handling that turns exposed endpoints into a full unauthenticated RCE surface via content-type confusion, with a public PoC already out. Prior research estimates the exposed n8n population at tens of thousands of hosts. The post-exploitation story mirrors Langflow. Workflows carry hardcoded API keys, database connection strings, and webhook secrets. RCE on the n8n host effectively gives access to every system the automations touch. vLLM vLLM is a high-throughput LLM serving engine with GPU acceleration, commonly used to self-host open-weight models in production. It exposes an OpenAI-compatible REST API. Authentication requires an explicit --api-key flag; without it, the API is open. The interesting finding from our scan was not vLLM itself but the adjacent deployments surfaced by the same query: OpenAI-compatible HTTP proxies, specifically LiteLLM-style gateways that aggregate multiple paid providers behind a single endpoint. These proxies store live API keys for OpenAI, Anthropic, Google, Groq, and Cohere. None had protection on the model list endpoint. One host exposed 35 models across multiple providers; several listed exclusively Anthropic Claude models. A proxy returns a model list only when the upstream provider authenticates, so every successful response confirms the underlying keys are live and billable. The abuse path is trivial: point any standard OpenAI SDK client at the proxy, enumerate the models, and, on hosts where prompt submission is also unprotected, send requests billed to the operator's accounts. It is the same credential-pivot pattern as Langflow and n8n. ComfyUI ComfyUI is a node-based workflow UI for Stable Diffusion, video generation, and multimodal image models. It runs on high-end GPU hardware with no authentication by default, making it a direct target for attackers looking to steal GPU compute. Our scan found open instances across a wide range of versions (v0.2.2 to 0.19.0), all of which were fully unauthenticated. The hardware exposure is the headline finding. Open hosts reported a combined ~4.3 TB of GPU VRAM, with cards ranging from RTX 4090s and RTX 5090s to datacenter-grade A100S and L40S units, each worth tens of thousands of dollars. An attacker can queue generation jobs against any of them at no cost. Beyond compute theft, 95% of open hosts expose a job history endpoint that leaks previously executed workflows, local file paths, and prior user content. About 12% advertise URL-loading nodes that act as server-side request forgery primitives: usable for internal network reconnaissance or cloud metadata credential theft. llama.cpp server llama-server is the HTTP server shipped with llama.cpp, commonly used to serve a single open-weight model in production. It has no authentication by default, no access controls on the inference endpoint, and a metadata endpoint that advertises exactly what the host is running. Anyone who reaches the port can submit prompts, watch active jobs, and burn the operator's GPU on their own workload. Classic LLMjacking, with the bonus of knowing exactly which model they are running. Of scanned hosts, 59% were confirmed open, more than any other platform in the scan. Everyone exposed its model name and hardware configuration, and about 37% also leaked real-time job state, confirming the host was actively serving users at the time of the scan. The models observed were standard open-weight builds rather than anything exotic, which is the point. An attacker is not looking for a rare model, just an unattended GPU. Jan Jan is an Electron desktop AI app with an optional OpenAI-compatible API server on port 1337. When enabled, it binds to all interfaces with no authentication. Jan is a useful example of how exposure surfaces unexpected content rather than how common it is. Our scan confirmed only two genuine Jan hosts. Both had gone offline by the rescan a week later. While one was live, it exposed a 35-model library that included miqu-70b; a leaked Mistral Medium prototype that was never officially released. When a desktop app binds its API server to the public internet, whatever model (or file path metadata) sits on the operator's disk becomes visible. Gradio Gradio is a Python framework for building ML demo apps: image classifiers, code interpreters, document Q&A, or anything a researcher can wrap in a web UI. Exposure risk depends entirely on what the underlying app does. A sentiment-analysis demo is low-stakes. An app that accepts file uploads, runs user code, or queries a database is a direct path to exploitation. The Gradio queue keeps processing submitted requests whether the operator is watching or not, so abuse can run quietly for days. Three unauthenticated bugs make unpatched instances worse: CVE-2024-47084: CORS validation bypass, fixed in 4.44.0; a malicious website can reach a locally running Gradio server while the victim is still logged in Ranking the Five Routes Each route carries operational trade-offs. The table below scores each on five dimensions, ranging from 0 (least favorable) to 5 (best for attacking): non-resistance (refusal behavior in response to offensive prompts), model capability (coding ability and parameter count), tool and MCP support, and effective token quota. Offensive LLMs score highest on non-resistance. But the underground-forum variants sit well below frontier models in capability and tool support, and subscriptions cap the quota. The crypto-middleman variant reaches frontier models via real provider accounts, but those accounts burn quickly once abuse is detected. Crypto payment for frontier models is for sure the best way to gain access for the most capable models with the ability to connect the model to any interface, such as MCPs, but it comes with some risks that the model might resist the action or the user will be blocked. Free-tier and keyless public APIs score well in capability and tool support, with full-function calling across most providers. The per-account quota is modest, tens of RPM, thousands of RPD, but trivial account rotation pushes the effective quota well above the face value. Stolen or leaked API keys, in principle, offer the best combination of capability and tool support; the retrohunt's 0.5% live rate shows the real-world quota is near zero. Exposed LLM servers score highest on non-resistance and token quota. Non-resistance is unconstrained: the attacker controls model selection, and our scan found at least one LM Studio host actively serving llama3.3-8b-instruct-thinking-heretic-uncensored-claude-4.5-opus-high-reasoning-i1. Token quota is equally unconstrained, bounded only by the victim's hardware rather than a billing cap. Capability and tool support vary by host, but that variance is what makes the route durable at scale. No individual host needs to run a frontier model. The scoring explains why exposed servers are the most durable route, even though they don't top every dimension. They are the only route where non-resistance and token quota both max out. The other three are each compromised on at least one of those two axes. Cases Found in the Wild Threat actors are now wiring malware to live LLM APIs, using them to generate malicious logic at runtime rather than embedding static code in the payload. Instead of scripting separate execution flows for different host conditions, the malware queries an LLM while running, determines whether the target appears to be a personal machine, a server, or an industrial controller, and then generates tailored commands or code accordingly. This shift matters because dynamically generated logic has no fixed signature to detect. Researchers have identified five malware families doing this. MalTerminal and PROMPTFLUX both use a hardcoded API key to connect to a commercial provider when needed. MalTerminal uses OpenAI GPT-4 via the now-retired chat-completions endpoint to create reverse shells or ransomware. PROMPTFLUX connects to Google gemini-1.5-flash-latest to rewrite its own VBScript source code between runs, making it harder to detect. LameHug, also known as PROMPTSTEAL, uses HuggingFace's Inference API to run Qwen 2.5-Coder-32B-Instruct for Windows commands to support reconnaissance and data theft. HuggingFace requires an API token for each request, but free accounts don't need a payment method and allow a few hundred requests per hour per API token. Attackers can easily create and rotate these API tokens, giving them the same access as stolen keys but with less hassle. PROMPTLOCK is a proof-of-concept AI-powered ransomware prototype, often called "Ransomware 3.0," developed by researchers at NYU's Tandon School of Engineering. The Go binary invokes gpt-oss-20b via a local Ollama API running on the infected host to generate Lua scripts that perform file listing, encryption, exfiltration, and (unfinished) wipe logic. This is a bring-your-own-model: no outbound calls, no provider-side billing trail, and no way to scale beyond the victim's own hardware. QUIETVAULT is a credential-theft variant. The JavaScript stealer exfiltrates GitHub and NPM tokens to an attacker-controlled GitHub repo and then hands off the filesystem search for additional secrets to whatever AI CLI is already installed on the victim, so the stolen credentials are an active on-host AI session rather than a bare API key. Looking at the four main routes discussed in this post, LameHug/PROMPTSTEAL is the best example of the free-tier method, since it calls HuggingFace's Inference API directly. MalTerminal and PROMPTFLUX both use hardcoded API keys, but it's unclear where those keys came from, so they could fit into the free-tier, crypto-middleman, or stolen-keys categories. QUIETVAULT is a twist on the stolen-credential method, using an on-host AI session instead of just a key. PROMPTLOCK is different because it uses a local model and only works on one victim at a time, so it doesn't fit into the four main routes and isn't discussed further. Across four routes: offensive LLMs for sale, free-tier and keyless public APIs, hardcoded keys in distributed artifacts, and exposed LLM servers on victim infrastructure, the most durable access is the last one. The precondition for abuse is almost never a sophisticated exploit. It is an unauthenticated port facing the internet. AI is the defining technology of this moment. It extends what a single person or small team can do and accelerates work that used to take weeks. AI is being integrated into more and more areas, from personal agents and email writing to some vulnerability research. The wow factor is real. But an LLM server is still a service running on a host. It listens on a port, speaks a protocol, and has an attack surface. The failure modes in this report; misconfiguration, leaked credentials, unpatched CVEs, open ports, are the same ones that produced years of incidents on Docker, Kubernetes, cloud storage, Redis, Elasticsearch, and bare Linux servers. The tooling is new. The mistakes are not. Two things follow. The operator is still responsible for the basics: authenticate the service, keep it off the public internet unless there is a reason to expose it, patch the known CVEs, and audit what is running. These are not AI-specific requirements. They are the same ones we have been making for every networked service. An exposed Ollama instance serving a stranger's prompts is not a failure of the model or the vendor that shipped it. It is a failure of whoever put it on the internet without a password. You broke it. You pay for it. IOCs Nicole Fishbein Nicole is a senior security researcher and malware analyst at Intezer. Prior to this, she was an embedded researcher in the IDF Intelligence Corps.

An information-stealing malware operation named Arkanix Stealer, promoted on multiple dark web forums towards the end of 2025, was likely developed as an AI-assisted experiment. The project included a control panel and a Discord server for communication with users, but the author took them down without notification, just two months after the operation began. Arkanix offered many of the standard data-stealing features that cybercriminals are used to, along with a modular architecture and anti-analysis features. Kaspersky researchers analyzed the Arkanix stealer and found clues indicating LLM-assisted development, which "might have drastically reduced development time and costs." The researchers believe that Arkanix was a short-lived project for quick financial gains, which makes detection and tracking much more difficult. Arkanix appears online Arkanix started being promoted on hacker forums in October 2025, offering two tiers to potential customers: a basic level with a Python-based implementation, and a “premium” one with a native C++ payload using VMProtect protection, integrating AV evasion and wallet injection features. The developer set up a Discord server that acted as a forum for the community around the project to receive updates, provide feedback for proposed features, and receive help. Also, a referral program was established to promote the project more aggressively, giving referrers an extra free hour of premium access, while potential new customers received one week of free access to the “premium” version. Data-stealing capabilities Arkanix malware can collect system information, steal data stored in the browser (history, autofill info, cookies, passwords), and cryptocurrency wallet data from 22 browsers. Kaspersky researchers say that it can also extract 0Auth2 tokens on Chromium-based browsers. Additionally, the malware can steal data from Telegram, steal Discord credentials, spread via the Discord API, and send messages to the victim’s friends/channels. Arkanix also targets credentials for Mullvad, NordVPN, ExpressVPN, and ProtonVPN, and can archive files from the local filesystem to exfiltrate them asynchronously. Additional modules that can be downloaded from the command-and-control include a Chrome grabber, a wallet patcher for Exodus or Atomic, a screenshots tool, HVNC, and stealers for FileZilla and Steam. The “premium” native C++ version adds RDP credential theft, anti-sandbox and anti-debugging checks, WinAPI-powered screen capturing, and also targets Epic Games, Battle.net, Riot, Unreal Engine, Ubisoft Connect, and GOG. The higher-tier variant also delivers the ChromElevator post-exploitation tool, which injects into suspended browser processes for data theft and is designed to bypass Google's App-Bound Encryption (ABE) protection for unauthorized access to user credentials. The purpose of the Arkanix stealer experiment remains unclear. The project may be an attempt to determine how LLM assistance can improve malware development and how quickly new features can be shipped to the community. Kaspersky's assessment is that Arkanix is "more of a public software product than a shady stealer." The researchers provide a comprehensive list of indicators of compromise (IoCs) that include hashes for detected files, along with domains and IP addresses. Overall prevention scores can hide what happens after initial access. Once attackers are using valid credentials, prevention drops sharply. The Blue Report 2026 measures defenses technique by technique across 338 million simulations run in customer production environments. Get the report

Introduction In October 2025, we discovered a series of forum posts advertising a previously unknown stealer, dubbed “Arkanix Stealer” by its authors. It operated under a MaaS (malware-as-a-service) model, providing users not only with the implant but also with access to a control panel featuring configurable payloads and statistics. The set of implants included a publicly available browser post-exploitation tool known as ChromElevator, which was delivered by a native C++ version of the stealer. This version featured a wide range of capabilities, from collecting system information to stealing cryptocurrency wallet data. Alongside that, we have also discovered Python implementation of the stealer capable of dynamically modifying its configuration. The Python version was often packed, thus giving the adversary multiple methods for distributing their malware. It is also worth noting that Arkanix was rather a one-shot malicious campaign: at the time of writing this article, the affiliate program appears to be already taken down. Kaspersky products detect this threat as Trojan-PSW.Win64.Coins.*, HEUR:Trojan-PSW.Multi.Disco.gen, Trojan.Python.Agent.*. Technical details Background In October 2025, a series of posts was discovered on various dark web forums, advertising a stealer referred to by its author as “Arkanix Stealer”. These posts detail the features of the stealer and include a link to a Discord server, which serves as the primary communication channel between the author and the users of the stealer. Upon further research utilizing public resources, we identified a set of implants associated with this stealer. Initial infection or spreading The initial infection vector remains unknown. However, based on some of the file names (such as steam_account_checker_pro_v1.py, discord_nitro_checker.py, and TikTokAccountBotter.exe) of the loader scripts we obtained, it can be concluded with high confidence that the initial infection vector involved phishing. Python loader The Python loader is the script responsible for downloading and executing the Python-based version of the Arkanix infostealer. We have observed both plaintext Python scripts and those bundled using PyInstaller or Nuitka, all of which share a common execution vector and are slightly obfuscated. These scripts often serve as decoys, initially appearing to contain legitimate code. Some of them do have useful functionality, and others do nothing apart from loading the stealer. Additionally, we have encountered samples that employ no obfuscation at all, in which the infostealer is launched in a separate thread via Python’s built-in threading module. Upon execution, the loader first installs the required packages — namely, requests, pycryptodome, and psutil — via the pip package manager, utilizing the subprocess module. On Microsoft Windows systems, the loader also installs pywin32. In some of the analyzed samples, this process is carried out twice. Since the loader does not perform any output validation of the module installation command, it proceeds to make a POST request to hxxps://arkanix[.]pw/api/session/create to register the current compromised machine on the panel with a predefined set of parameters even if the installation failed. After that, the stealer makes a GET request to hxxps://arkanix[.]pw/stealer.py and executes the downloaded payload. Python stealer version During our research, we obtained a sample of the Python implementation of the Arkanix stealer, which was downloaded from the endpoint hxxps://arkanix[.]pw/stealer.py by the previous stage. The stealer’s capabilities — or features, as referred to by the author — in this version are configurable, with the default configuration predefined within the script file. To dynamically update the feature list, the stealer makes a GET request to hxxps://arkanix[.]pw/api/features/{payload_id}, indicating that these capabilities can be modified on the panel side. The feature list is identical to the one that was described in the GDATA report. Prior to executing the information retrieval-related functions, the stealer makes a request to hxxps://arkanix[.]pw/upload_dropper.py, saves the response to %TEMP%\upd_{random 8-byte name}.py, and executes it. We do not have access to the contents of this script, which is referred to as the “dropper” by the attackers. During its main information retrieval routine, at the end of each processing stage, the collected information is serialized into JSON format and saved to a predefined path, such as %LOCALAPPDATA\Arkanix_lol\%info_class%.json. In the following, we will provide a more detailed description of the Python version’s data collection features. System info collection Arkanix Stealer is capable of collecting a set of info about the compromised system. This info includes: OS version CPU and GPU info RAM size Screen resolution Keyboard layout Time zone Installed software Antivirus software VPN Information collection is performed using standard shell commands with the exception of the VPN check. The latter is implemented by querying the endpoint hxxps://ipapi[.]co/json/ and verifying whether the associated IP address belongs to a known set of VPNs, proxies, or Tor exit nodes. Browser features This stealer is capable of extracting various types of data from supported browsers (22 in total, ranging from the widely popular Google Chrome to the Tor Browser). The list of supported browsers is hardcoded, and unlike other parameters, it cannot be modified during execution. In addition to a separate Chrome grabber module (which we’ll discuss later), the stealer itself supports the extraction of diverse information, such as: Browser history (URLs, visit count and last visit) Autofill information (email, phone, addresses and payment cards details) Saved passwords Cookies In case of Chromium-based browsers, 0Auth2 data is also extracted All information is decrypted using either the Windows DPAPI or AES, where applicable, and searched for relevant keywords. In the case of browser information collection, the stealer searches exclusively for keywords related to banking (e.g., “revolut”, “stripe”, “bank”) and cryptocurrencies (e.g., “binance”, “metamask”, “wallet”). In addition to this, the stealer is capable of extracting extension data from a hardcoded list of extensions associated with cryptocurrencies. Telegram info collection Telegram data collection begins with terminating the Telegram.exe process using the taskkill command. Subsequently, if the telegram_optimized feature is set to False, the malware zips the entire tdata directory (typically located at %APPDATA%\Roaming\Telegram Desktop\tdata) and transmits it to the attacker. Otherwise, it selectively copies and zips only the subdirectories containing valuable info, such as message log. The generated archive is sent to the endpoint /delivery with the filename tdata_session.zip. Discord capabilities The stealer includes two features connected with Discord: credentials stealing and self-spreading. The first one can be utilized to acquire credentials both from the standard client and custom clients. If the client is Chromium-based, the stealer employs the same data exfiltration mechanism as during browser credentials stealing. The self-spreading feature is configurable (meaning it can be disabled in the config). The stealer acquires the list of user’s friends and channels via the Discord API and sends a message provided by the attacker. This stealer does not support attaching files to such messages. VPN data collection The VPN collector is searching for a set of known VPN software to extract account credentials from the credentials file with a known path that gets parsed with a regular expression. The extraction occurs from the following set of applications: Mullvad VPN NordVPN ExpressVPN ProtonVPN File retrieval File retrieval is performed regardless of the configuration. The script relies on a predefined set of paths associated with the current user (such as Desktop, Download, etc.) and file extensions mainly connected with documents and media. The script also has a predefined list of filenames to exfiltrate. The extracted files are packed into a ZIP archive which is later sent to the C2 asynchronously. An interesting aspect is that the filename list includes several French words, such as “motdepasse” (French for “password”), “banque” (French for “bank”), “secret” (French for “secret”), and “compte” (French for “account”). Other payloads We were able to identify additional modules that are downloaded from the C2 rather than embedded into the stealer script; however, we weren’t able to obtain them. These modules can be described by the following table, with the “Details” column referring to the information that could be extracted from the main stealer code. The Wallet patcher and Extra collector scripts are received in an encrypted form from the C2 server. To decrypt them, the attackers utilize the AES-GCM algorithm in conjunction with PBKDF2 (HMAC and SHA256). After decryption, the additional payload has its template placeholders replaced and is stored under a partially randomized name within a temporary folder. Once all operations are completed, the stealer removes itself from the drive, along with the artifacts folder (Arkanix_lol in this case). Native version of stealer During our analysis, we were able to obtain both the release and debug versions of the native implementation, as both were uploaded to publicly available resources. The following are the key differences between the two: The release version employs VMProtect, but does not utilize code virtualization. The debug version communicates with a Discord bot for command and control (C2), whereas the release version uses the previously mentioned C2 domain arkanix[.]pw . The debug version includes extensive logging, presumably for the authors’ debugging purposes. Notably, the native implementation explicitly references the name of the stealer in the VersionInfo resources. This naming convention is consistent across both the debug version and certain samples containing the release version of the implant. After launching, the stealer implements a series of analysis countermeasures to verify that the application is not being executed within a sandboxed environment or run under a debugger. Following these checks, the sample patches AmsiScanBuffer and EtwEventWrite to prevent the triggering of any unwanted events by the system. Once the preliminary checks are completed, the sample proceeds to gather information about the system. The list of capabilities is hardcoded and cannot be modified from the server side, in contrast to the Python version. What is more, the feature list is quite similar to the Python version except a few ones. RDP connections The stealer is capable of collecting information about known RDP connections that the compromised user has. To achieve this, it searches for .rdp files in %USERPROFILE%\Documents and extracts the full server address, password, username and server port. Gaming files The stealer also targets gamers and is capable to steal credentials from the popular gaming platform clients, including: Steam Epic Games Launcher net Riot Origin Unreal Engine Ubisoft Connect GOG Screenshots The native version, unlike its Python counterpart, is capable of capturing screenshots for each monitor via capCreateCaptureWindowA WinAPI. In conclusion, this sample communicates with the C2 server through the same endpoints as the Python version. However, in this instance, all data is encrypted using the same AES-GCM + PBKDF2 (HMAC and SHA256) scheme as partially employed in the Python variant. In some observed samples, the key used was arkanix_secret_key_v20_2024. Alongside that, the C++ sample explicitly sets the User-Agent to ArkanixStealer/1.0. Post-exploitation browser data extractor This is an implant embedded within the resources of the C++ implementation. The author incorporated it into the resource section without applying any obfuscation or encryption. Subsequently, the stealer extracts the payload to a temporary folder with a randomly generated name composed of hexadecimal digits (0-9 and A-F) and executes it using the CreateProcess WinAPI. The payload itself is the unaltered publicly available project known as “ChromElevator”. To summarize, this tool consists of two components: an injector and the main payload. The injector initializes a direct syscall engine, spawns a suspended target browser process, and injects the decrypted code into it via Nt syscalls. The injected payload then decrypts the browser master key and exfiltrates data such as cookies, login information, web data, and so on. Infrastructure During the Arkanix campaign, two domains used in the attacks were identified. Although these domains were routed through Cloudflare, a real IP address was successfully discovered for one of them, namely, arkanix[.]pw. For the second one we only obtained a Cloudflare IP address. Both servers were also utilized to host the stealer panel, which allows attackers to monitor their victims. The contents of the panel are secured behind a sign-in page. Closer to the end of our research, the panel was seemingly taken down with no message or notice. Stealer promotion During the research of this campaign, we noticed that the forum posts advertising the stealer contained a link leading to a Discord server dubbed “Arkanix” by the authors. The server posed as a forum where authors posted various content and clients could ask various questions regarding this malicious software. While users mainly thank and ask about when the feature promised by the authors will be released and added into the stealer, the content made by the authors is broader. The adversary builds up the communication with potential buyers using the same marketing and communication methods real companies employ. To begin with, they warm up the audience by posting surveys about whether they should implement specific features, such as Discord injection and binding with a legitimate application (sic!). Additionally, the author promised to release a crypter as a side project in four to six weeks, at the end of October. As of now, the stealer seems to have been taken down without any notice while the crypter was never released. Furthermore, the Arkanix Stealer authors decided to implement a referral program to attract new customers. Referrers were promised an additional free hour to their premium license, while invited customers received seven days of free “premium” trial use. As stated in forum posts, the premium plan included the following features: C++ native stealer Exodus and Atomic cryptocurrency wallets injection Increased payload generation, up to 10 payloads Priority support Speaking of technical details, based on the screenshot of the Visual Studio stealer project that was sent to the Discord server, we can conclude that the author is German-speaking. This same screenshot also serves as a probable indicator of AI-assisted development as it shares the common patterns of such assistants, e.g. the presence of the utils.cpp file. What provides even more confidence is the overall code structure, the presence of comments and extensive debugging log output. Conclusions Information stealers have always posed as a serious threat to users’ data. Arkanix is no exception as it targets a wide range of users, from those interested in cryptocurrencies and gaming to those using online banking. It collects a vast amount of information including highly sensitive personal data. While being quite functional, it contains probable traces of LLM-assisted development which suggests that such assistance might have drastically reduced development time and costs. Hence it follows that this campaign tends to be more of a one-shot campaign for quick financial gains rather than a long-running infection. The panel and the Discord chat were taken down around December 2025, leaving no message or traces of further development or a resurgence. In addition, the developers behind the Arkanix Stealer decided to address the public, implementing a forum where they posted development insights, conducted surveys and even ran a referral program where you could get bonuses for “bringing a friend”. This behavior makes Arkanix more of a public software product than a shady stealer. Indicators of Compromise Additional IoCs are available to customers of our Threat Intelligence Reporting service. For more details, contact us at [email protected]. Domains and IPs arkanix[.]pw arkanix[.]ru

Hackers have released stolen data belonging to US insurance giant Allianz Life, exposing 2.8 million records with sensitive information on business partners and customers in ongoing Salesforce data theft attacks. Last month, Allianz Life disclosed that it suffered a data breach when the personal information for the "majority" of its 1.4 million customers was stolen from a third-party, cloud-based CRM system on July 16th. While the company did not name the provider, BleepingComputer first reported the incident was part of a wave of Salesforce-targeted thefts carried out by the ShinyHunters extortion group. Over the weekend, ShinyHunters and other threat actors claiming overlap with "Scattered Spider" and "Lapsus$" created a Telegram channel called "ScatteredLapsuSp1d3rHunters" to taunt cybersecurity researchers, law enforcement, and journalists while taking credit for a string of high-profile breaches. Many of these attacks had not previously been attributed to any threat actor, including the attacks on Internet Archive, Pearson, and Coinbase. One of the attacks claimed by the threat actors is Allianz Life, for which they proceeded to leak the complete databases that were stolen from the company's Salesforce instances. These files consist of the Salesforce "Accounts" and "Contacts" database tables, containing approximately 2.8 million data records for individual customers and business partners, such as wealth management companies, brokers, and financial advisors. The leaked Salesforce data includes sensitive personal information, such as names, addresses, phone numbers, dates of birth, and Tax Identification Numbers, as well as professional details like licenses, firm affiliations, product approvals, and marketing classifications. BleepingComputer has been able to confirm with multiple people that their data in the leaked files is accurate, including their phone numbers, email addresses, tax IDs, and other information contained in the database. BleepingComputer contacted Allianz Life about the leaked database but was told that they could not comment as the investigation is ongoing. The Salesforce data-theft attacks The Salesforce data theft attacks are believed to have started at the beginning of the year, with the threat actors conducting social engineering attacks to trick employees into linking a malicious OAuth app with their company's Salesforce instances. Once linked, the threat actors used the connection to download and steal the databases, which were then used to extort the company through email. Extortion demands were sent to the companies via email and were signed as coming from ShinyHunters. This notorious extortion group has been linked to many high-profile attacks over the years, including those against AT&T, PowerSchool, and the SnowFlake attacks. While ShinyHunters is known to target cloud SaaS applications and website databases, they are not known for these types of social engineering attacks, causing many researchers and the media to attribute some of the Salesforce attacks to Scattered Spider. However, ShinyHunters told BleepingComputer the "ShinyHunters" group and "Scattered Spider" are now one and the same. "Like we have said repeatedly already, ShinyHunters and Scattered Spider are one and the same," ShinyHunters told BleepingComputer. "They provide us with initial access and we conduct the dump and exfiltration of the Salesforce CRM instances. Just like we did with Snowflake." It is also believed that many of the group's members share their roots in another hacking group known as Lapsus$, which was responsible for numerous attacks in 2022-2023, before some of their members were arrested. Lapsus$ was behind breaches at Rockstar Games, Uber, 2K, Okta, T-Mobile, Microsoft, Ubisoft, and NVIDIA. Like Scattered Spider, Lapsus$ was also adept at social engineering attacks and SIM swap attacks, allowing them to run over billion and trillion-dollar companies' IT defenses. Over the past couple of years, there have been many arrests linked to all three collectives, so it's not clear if the current threat actors are old threat actors, new ones who have picked up the mantle, or are simply utilizing these names to plant false flags. Overall prevention scores can hide what happens after initial access. Once attackers are using valid credentials, prevention drops sharply. The Blue Report 2026 measures defenses technique by technique across 338 million simulations run in customer production environments. Get the report