{"description":"Trending threats, MITRE ATT\u0026CK coverage, and detection metadata. Fed continuously.","feed_url":"https://feed.craftedsignal.io/vendors/nomic-ai/feed.json","home_page_url":"https://feed.craftedsignal.io/","items":[{"_cs_actors":[],"_cs_cpes":[],"_cs_cves":[],"_cs_exploited":false,"_cs_has_poc":false,"_cs_poc_references":[],"_cs_products":["Claude","ChatGPT","Codex","Ollama","LM Studio","GPT4All","huggingface.co","jan.ai","nomic.ai","replicate.com","civitai.com","KoboldAI","Oobabooga Text Generation Web UI","modelscope.cn","Dashscope","Tongyi","openrouter.ai"],"_cs_severities":["medium"],"_cs_tags":["local-llm","shadow-ai","dns-monitoring","data-exfiltration","policy-violation","windows","endpoint-security"],"_cs_type":"advisory","_cs_vendors":["Anthropic","OpenAI","Ollama","LM Studio","Hugging Face","Nomic AI","Replicate","Alibaba Cloud","OpenRouter"],"content_html":"\u003cp\u003eThis threat brief outlines the detection of DNS queries made by local Large Language Model (LLM) frameworks operating on Windows endpoints. Organizations are increasingly facing challenges with \u0026quot;shadow AI\u0026quot; as employees independently deploy local LLM tools such as Ollama, LM Studio, and GPT4All. These frameworks often make DNS calls to various external repositories and services, including huggingface.co, ollama.ai, and openai.com, for model downloads, updates, and telemetry. Such network activity, captured via Sysmon DNS query events (Event ID 22), can indicate unauthorized software usage, potential policy violations, and, critically, data exfiltration risks if sensitive corporate information is processed by these unmanaged models or sent out through their telemetry. Detecting these queries is vital for maintaining network security, protecting intellectual property, and ensuring compliance within corporate environments.\u003c/p\u003e\n\u003ch2 id=\"impact\"\u003eImpact\u003c/h2\u003e\n\u003cp\u003eThe unauthorized use of local LLM frameworks within an enterprise environment poses several significant risks. It can lead to the processing of sensitive corporate data by unvetted and unmanaged applications, creating avenues for inadvertent data leakage or malicious exfiltration through model telemetry or update mechanisms. Such activities also bypass established security controls and introduce unmonitored software, contributing to \u0026quot;shadow IT\u0026quot; environments that are difficult to secure and audit. This can result in intellectual property theft, violations of regulatory compliance, and a general erosion of data governance policies, potentially causing financial damages and reputational harm to the organization.\u003c/p\u003e\n\u003ch2 id=\"recommendation\"\u003eRecommendation\u003c/h2\u003e\n\u003cul\u003e\n\u003cli\u003eEnable Sysmon process creation and DNS query logging (Event ID 22) across all Windows endpoints to ensure the necessary telemetry for this detection is collected.\u003c/li\u003e\n\u003cli\u003eDeploy the \u0026quot;Detect Local LLM Framework DNS Queries\u0026quot; Sigma rule to your SIEM and tune it for your specific environment, creating exceptions for authorized LLM usage by developers or sanctioned AI/ML workstations.\u003c/li\u003e\n\u003cli\u003eBlock the IOCs listed in this brief at your network perimeter (e.g., DNS resolver, proxy, firewall) to prevent unauthorized access to known LLM model repositories and services.\u003c/li\u003e\n\u003cli\u003eReview network logs and endpoint telemetry for any activity matching the DNS query patterns identified in the \u0026quot;Detect Local LLM Framework DNS Queries\u0026quot; rule and investigate originating processes.\u003c/li\u003e\n\u003c/ul\u003e\n","date_modified":"2026-07-27T18:18:49Z","date_published":"2026-07-27T18:18:49Z","id":"https://feed.craftedsignal.io/briefs/2026-07-local-llm-framework-dns-query/","summary":"This brief details the detection of DNS queries originating from local Large Language Model (LLM) frameworks like Ollama, LM Studio, and GPT4All on Windows endpoints, leveraging Sysmon Event ID 22 to identify potential unauthorized AI tool usage or data exfiltration risks associated with model downloads, updates, and telemetry from repositories such as huggingface.co and ollama.ai.","title":"Detection of Local LLM Framework DNS Queries","url":"https://feed.craftedsignal.io/briefs/2026-07-local-llm-framework-dns-query/"},{"_cs_actors":[],"_cs_cpes":[],"_cs_cves":[],"_cs_exploited":false,"_cs_has_poc":false,"_cs_poc_references":[],"_cs_products":["Ollama","llama.cpp","GPT4All","LM Studio","Jan AI","KoboldCPP","NutStudio","Oobabooga","HuggingFace Transformers","LangChain"],"_cs_severities":["medium"],"_cs_tags":["shadow-it","llm","data-exfiltration","policy-violation","endpoint","shadow-ai","local-llm","intellectual-property-theft","endpoint-security","windows"],"_cs_type":"advisory","_cs_vendors":["Ollama","GGUF.ai","Nomic AI","Jan","Hugging Face","LangChain"],"content_html":"\u003cp\u003eThis threat brief focuses on the detection of Large Language Model (LLM) file creation on Windows endpoints, a key indicator of potential \u0026quot;shadow AI\u0026quot; deployments or unauthorized local LLM infrastructure. Organizations are facing increasing risks from employees downloading and running open-source or local LLMs on corporate devices, often bypassing established security controls and data governance policies. These local inference frameworks, such as Ollama, llama.cpp, GPT4All, and LM Studio, utilize specific file formats like quantized models (.gguf, .ggml), safetensors files, and proprietary Modelfiles. The creation of these files suggests that LLMs are being run locally, potentially processing sensitive company data outside of approved, monitored environments. This can lead to significant data exfiltration risks, intellectual property leakage, and security blind spots, as these decentralized AI deployments often operate without the oversight of enterprise monitoring systems, making it difficult to track data usage and ensure compliance. This detection method provides visibility into such activities, enabling defenders to identify and mitigate these risks proactively.\u003c/p\u003e\n\u003ch2 id=\"attack-chain\"\u003eAttack Chain\u003c/h2\u003e\n\u003cp\u003eThis brief describes a detection for the presence of unauthorized Large Language Model (LLM) files, rather than a specific attack chain or exploitation scenario. The detection focuses on identifying the creation of specific file types associated with local LLM inference engines. While the creation of these files is not an \u0026quot;attack\u0026quot; in itself, it indicates the establishment of a local environment that can be leveraged for various malicious or policy-violating activities. Therefore, a traditional multi-step attack chain, from initial access to impact, is not directly applicable or described in the source material for this specific detection.\u003c/p\u003e\n\u003ch2 id=\"impact\"\u003eImpact\u003c/h2\u003e\n\u003cp\u003eThe primary impact of undetected local LLM model file creation is the potential for significant data exfiltration and intellectual property loss. Employees using unapproved LLMs on corporate devices might inadvertently or intentionally feed sensitive company data into these models, leading to data breaches or compliance violations. The proliferation of shadow AI infrastructure creates severe security blind spots, making it challenging for security teams to monitor data flows, detect malicious activity, and ensure adherence to organizational policies. Furthermore, these rogue LLM deployments can consume significant system resources, impacting legitimate business operations, and introduce new attack vectors if the downloaded models or frameworks contain vulnerabilities or malicious code. Without proper governance, organizations face increased legal, reputational, and financial risks due to unmanaged AI usage.\u003c/p\u003e\n\u003ch2 id=\"recommendation\"\u003eRecommendation\u003c/h2\u003e\n\u003cul\u003e\n\u003cli\u003eDeploy the Sigma rule \u003ccode\u003eDetect_Local_LLM_Model_File_Creation\u003c/code\u003e to your SIEM solution to identify instances of LLM model file creation.\u003c/li\u003e\n\u003cli\u003eEnsure Sysmon Event ID 11 (File Creation) logging is enabled across all Windows endpoints to provide the necessary telemetry for the rule.\u003c/li\u003e\n\u003cli\u003eRegularly review alerts generated by \u003ccode\u003eDetect_Local_LLM_Model_File_Creation\u003c/code\u003e and investigate the context of LLM file creation to determine if it is authorized or represents a policy violation.\u003c/li\u003e\n\u003cli\u003eEducate employees about the risks associated with unauthorized local LLM usage and the proper channels for AI tool adoption to mitigate data exfiltration risks and policy violations.\u003c/li\u003e\n\u003c/ul\u003e\n","date_modified":"2026-07-27T18:26:33Z","date_published":"2026-07-27T18:17:13Z","id":"https://feed.craftedsignal.io/briefs/2026-07-llm-model-file-creation/","summary":"This brief describes how the creation of Large Language Model (LLM) files, including formats like .gguf, .safetensors, .ggml, and Modelfiles, by local AI inference frameworks such as Ollama, llama.cpp, GPT4All, and LM Studio can be detected on Windows endpoints, indicating potential shadow AI deployments, unauthorized model downloads, or rogue LLM infrastructure which poses data exfiltration risks and policy violations.","title":"Detection of Local LLM Model File Creation on Endpoints","url":"https://feed.craftedsignal.io/briefs/2026-07-llm-model-file-creation/"}],"language":"en","title":"CraftedSignal Threat Feed - Nomic AI","version":"https://jsonfeed.org/version/1.1"}