The same Cornell analysis cited a Cisco study using algorithmic jailbreaking. DeepSeek failed to block all 50 harmful prompts in that evaluation, while ChatGPT blocked 86 percent. That finding concerns the tested systems and methods. It does not establish that every use of either model will produce the same result.<\/p>\n
Research published in DeepSeek for healthcare: do no harm?<\/a> reported that DeepSeek models had been downloaded more than 10 million times from Hugging Face and had more than 57 million downloads from the Apple and Google app stores, based on May 2025 data. The research also noted reported training costs of $5.576 million for the official training of DeepSeek-v3, with performance reported as similar to Llama 3.1, GPT-4o, o1 and Claude Sonnet on common large-language-model benchmarks. These figures describe reported results in that research and should not be treated as a guarantee for a particular business deployment.<\/p>\n Academic interest has been positive in some fields. A mini review in Frontiers in Artificial Intelligence<\/a> analyzed research containing DeepSeek in its title, abstract or keywords and found generally positive sentiment in the included English-language literature, with computer science showing the highest mean sentiment. The review also said that discussion of political and ethical implications was limited in the academic literature it examined.<\/p>\n The federal order applies to civilian agencies, so a private company should not describe it as a direct prohibition on all businesses. It does provide a useful checklist for reviewing how AI tools enter and leave an office environment. The most relevant issue is whether a service can receive company data and whether the IT team knows what technology is behind it.<\/p>\n Start with the software inventory. Include desktop and mobile applications, browser-based AI tools, coding assistants, document-processing services and any customer-facing chatbot. Record whether each product is hosted by the vendor, self-hosted by the business or connected to an internal system. A service may look like a general productivity tool while using a third-party model or API underneath.<\/p>\n Next, review contracts and vendor disclosures. The federal order specifically includes third-party services developed using DeepSeek technology, which is the practical reason to look beyond app names. Ask vendors what models power the service, where information is processed, what information is retained and whether the vendor can identify a model or subprocess changes in a later update. These are review questions, not claims about any named vendor.<\/p>\n Then examine permissions and data paths. Check whether staff can paste customer records, employee details, contracts, credentials or source code into an AI service. Review account access, administrator controls and logging. If a business cannot establish where information sent to a service is stored or processed, that gap should be resolved before the service is approved for sensitive work.<\/p>\n Local deployment is one factor in this review, but it is not an automatic answer to every risk. The Cornell analysis said that running DeepSeek offline can reduce the risk of data leakage. A 2026 Frontiers in Artificial Intelligence<\/a> review also described self-hosting as a way for organizations to control data governance and customization, while noting that the model is not fully open source because its training data have not been made entirely available. A local setup still needs access controls, patching and a decision about which information staff may submit.<\/p>\n DeepSeek’s growing use does not, by itself, show that a business has a security incident. It does show that AI procurement needs the same attention as other software dependencies. The main questions are which model is in use, who operates it, what data enters it and whether an outside company can change the technology without the customer seeing that change.<\/p>\n The federal action is significant because it treats a model’s origin and supply chain as part of the procurement decision. The order also affects third-party services, a detail that matters to small businesses because AI functionality can be added by a vendor without appearing in a simple application inventory. A documented review gives the IT team a way to identify those dependencies before they become unclear.<\/p>\n On October 19, 2026, the Department of Homeland Security ordered all federal civilian agencies to remove DeepSeek products and services from their systems and stop buying them, citing national security risks. The order was published in the Federal Register on October 20, 2026.<\/p>\n DHS warns that DeepSeek’s models present \u201cmultiple threat vectors\u201d that could give the Chinese Communist Party access to U.S. government systems and allow sensitive data to be transferred to China. The department also raised concerns about tracking or digitally fingerprinting users.<\/p>\n DHS invoked the 2018 Federal Acquisition Supply Chain Security Act following a review by the Federal Acquisition Security Council. The order appears as a five-page Federal Register notice, document 2026-21447, at 91 FR 66346.<\/p>\n BizScoreAI scores how visible a business is to AI search and shows what its listing looks like to the engines people ask. Open the free listing check<\/a>.<\/p>\nWhat should a small business IT team check?<\/h2>\n
What to do now<\/h2>\n
\n
What this means for private businesses<\/h2>\n
FAQ<\/h2>\n
What did the Trump administration order regarding DeepSeek?<\/h3>\n
Why is DeepSeek being banned from federal agencies?<\/h3>\n
What legal authority did DHS use to ban DeepSeek?<\/h3>\n
BizScoreAI<\/h2>\n
![\"BizScoreAI,](\"https:\/\/bizscoreai.com\/wp-content\/uploads\/2026\/04\/BizScoreAI-Social-image.jpg\")
<\/a><\/p>\n