Artificial intelligence is rapidly transforming how governments and businesses operate. From automating administrative tasks to analyzing large volumes of data, AI tools are becoming part of everyday workflows. However, as organizations adopt these technologies, questions about data privacy, national security, vendor trust, and regulatory compliance are becoming increasingly important.
A new dispute between the United States and China over the AI company DeepSeek highlights these challenges. The US Department of Homeland Security has ordered federal civilian agencies to remove DeepSeek products and related third-party solutions built using DeepSeek technology from their IT systems. The order also restricts agencies from procuring these products and services.
China has strongly opposed the decision, arguing that the United States is using national security concerns to restrict Chinese technology companies.
Beyond the geopolitical disagreement, the situation offers an important lesson for organizations everywhere: AI adoption must be accompanied by clear security controls, data governance, and risk-based vendor assessments.
What Happened?
On October 19, 2026, the US Department of Homeland Security published a notice concerning restrictions on DeepSeek products and services across federal civilian government systems.
The order, issued earlier in July, covers products developed by DeepSeek and its affiliated entities. It also extends to third-party products and services developed using DeepSeek implementations. The restrictions address both the removal of affected products from agency IT environments and the procurement of such technologies.
The US government cited several potential security concerns, including the possibility that sensitive information entered into these systems could become accessible to Chinese authorities, depending on how the services operate and applicable legal requirements. The notice also raised concerns about potential security vulnerabilities and other risks associated with the models.
The restrictions do not automatically mean that every DeepSeek model has been proven malicious or that every use of the technology results in data exposure. Rather, the government action reflects its assessment of the risks associated with deploying these products within federal civilian agencies.
DeepSeek’s models are widely available, including through open-weight releases, making the issue more complex than simply removing a single application. Organizations may also need to identify third-party services, internal applications, and development workflows that incorporate the technology.
How Has China Responded?
Chinese officials criticized the US decision and described it as an unjustified use of national security arguments to limit competition in technology and trade.
China’s Foreign Ministry said the country would defend the legitimate interests of its companies. Its Ministry of Commerce also expressed strong opposition and warned that it would monitor the situation and consider measures to protect those interests.
The dispute comes amid broader competition between the two countries over AI development, advanced semiconductor technology, intellectual property, and access to strategic computing resources.
Although both countries have explored areas of cooperation on AI governance, they continue to disagree over technology access, security risks, and the role of government in regulating emerging AI systems.
For businesses, the implications extend beyond the political debate. Technology restrictions can affect procurement decisions, software dependencies, service availability, and the long-term viability of AI systems built around particular providers.
Why Is DeepSeek Raising AI Security Concerns?
The central issue is not simply where an AI company is headquartered. It is how an AI system collects, processes, stores, and transfers data, what access its provider may have, and whether the system can be evaluated and governed appropriately.
Several security considerations deserve attention.
1. Sensitive Data Exposure
Employees may enter confidential business information, customer records, source code, internal documents, or government-related information into AI tools.
If an organization does not understand the service’s data handling practices, it may expose information to unauthorized parties or violate contractual and regulatory obligations.
Before approving an AI service, organizations should examine data retention, processing locations, access permissions, training-data policies, and options for deleting submitted information.
2. Third-Party and Supply Chain Risk
AI products are increasingly embedded in applications developed by other vendors. A business might not directly subscribe to DeepSeek but could use a software product that integrates one of its models.
This creates a broader supply chain challenge. Security teams must understand not only which AI tools employees use, but also which models and external services are embedded in the organization’s technology stack.
A reliable software inventory and vendor assessment process can help identify these dependencies before they become operational or compliance problems.
3. Model and Application Security
AI models can introduce risks beyond traditional software vulnerabilities. Depending on how they are integrated, potential issues include prompt injection, insecure tool access, excessive permissions, sensitive data leakage, and unsafe outputs.
Organizations should test the complete AI application, including its interfaces, retrieval systems, plugins, APIs, and connections to internal resources.
A model that performs well in demonstrations should not automatically be considered safe for production use.
4. Regulatory and Data Governance Challenges
AI services may process personal information, confidential business data, or regulated records. Their use can therefore affect obligations under privacy laws, industry regulations, contractual requirements, and internal security policies.
Organizations need documented rules governing approved AI tools, permitted data types, user access, retention, monitoring, and incident reporting.
Compliance should be assessed according to the organization’s jurisdiction, industry, data, and actual deployment model rather than assuming that a single AI policy satisfies every requirement.
What This Means for Businesses Beyond the United States
The US restriction applies to federal civilian government systems and procurement covered by the order. It should not be interpreted as a blanket prohibition on DeepSeek for every private organization or individual worldwide.
Nevertheless, businesses in other countries can learn from the situation.
Organizations in India, Europe, North America, and other regions should assess AI tools according to their own threat models, contractual obligations, applicable laws, and business requirements.
For example, an organization may choose to use a particular model in a controlled environment while prohibiting the submission of confidential information to externally hosted services. Another organization handling highly sensitive government or healthcare data may require stricter restrictions or an alternative deployment architecture.
The right decision depends on evidence, risk tolerance, and the sensitivity of the information involved.
Practical Steps to Secure Enterprise AI Adoption
Organizations can reduce exposure by taking the following measures.
1. Establish an approved AI inventory
Identify AI tools used by employees, development teams, business units, and third-party applications. Include locally hosted models, external APIs, AI-enabled SaaS products, and embedded model dependencies.
2. Classify information before it enters AI systems
Define which data can be submitted to approved tools. Restrict or prohibit the use of sensitive personal information, credentials, confidential source code, and regulated records unless the environment has been specifically assessed and authorized.
3. Evaluate AI vendors and their data practices
Review data processing locations, retention policies, security controls, subprocessors, incident notification procedures, and contractual commitments. Reassess vendors when their services or risk profiles change.
4. Apply least-privilege access
AI applications should receive only the permissions necessary to perform their intended tasks. Avoid giving models unrestricted access to enterprise databases, file systems, internal applications, or administrative functions.
5. Test AI systems before deployment
Conduct security assessments that address prompt injection, data leakage, unsafe tool execution, access control weaknesses, API vulnerabilities, and other risks relevant to the implementation.
6. Monitor usage and investigate unusual activity
Maintain appropriate logs of AI access, model interactions, tool calls, and data transfers. Monitoring should respect privacy requirements while helping security teams identify policy violations and suspicious behavior.
7. Develop an AI incident response plan
Prepare for scenarios such as accidental disclosure of confidential information, compromised AI integrations, unauthorized model access, or the discovery that a deployed vendor is no longer approved.
The plan should identify who can disable a service, investigate the incident, assess regulatory obligations, and communicate with affected stakeholders.
8. Review compliance continuously
AI governance is not a one-time approval process. Organizations should periodically reassess vendors, model versions, data flows, security controls, and applicable legal requirements.
Industries That Should Pay Attention
Financial Services
Banks, insurers, and payment providers should evaluate AI tools that may process customer information, transaction data, fraud intelligence, or internal financial records. Strong access controls, vendor assessments, and data loss prevention measures are essential.
Healthcare and Life Sciences
Hospitals, laboratories, and healthcare technology companies should carefully assess AI tools that may handle patient records, clinical information, or protected health information. Data governance, access restrictions, and compliance reviews should be part of the approval process.
Retail and E-commerce
Retailers should review AI tools used for customer service, personalization, inventory planning, and customer analytics. Security controls should prevent unnecessary exposure of customer information and ensure that third-party integrations are appropriately governed.
Manufacturing
Manufacturers may use AI to support product design, predictive maintenance, supply chain operations, and industrial analytics. Organizations should protect proprietary designs, operational information, and connected systems while limiting unnecessary model access.
Government and Public Sector
Government agencies should evaluate AI tools against their procurement requirements, information classification rules, data residency obligations, and national security policies. Where specific restrictions apply, agencies and contractors should verify their scope before deploying affected technology.
Technology and SaaS Companies
Software providers should maintain visibility into models used by their development tools and customer-facing products. Secure development practices, model evaluations, dependency reviews, and continuous monitoring can help reduce supply chain and application risks.
Conclusion
The dispute over DeepSeek illustrates how AI security, national policy, and technology supply chains are becoming increasingly interconnected.
For governments, the priority is protecting sensitive information and managing risks associated with strategic technology dependencies. For businesses, the challenge is determining which AI tools can be used safely, under what conditions, and with which safeguards.
The key lesson is not to approve or reject an AI system solely because of its country of origin. Organizations should make informed decisions based on data handling practices, security testing, transparency, access controls, legal requirements, and the consequences of a potential compromise.
As AI becomes embedded in more business processes, organizations need a structured approach to AI governance that connects cybersecurity, privacy, procurement, compliance, and operational risk management.
Responsible AI adoption starts with knowing what you use, understanding what data it can access, and verifying that appropriate security controls are in place.
About COE Security
COE Security partners with organizations in financial services, healthcare, retail, manufacturing, and government to secure AI-powered systems and ensure compliance.
Our offerings include:
- AI-enhanced threat detection and real-time monitoring
- Data governance aligned with GDPR, HIPAA, and PCI DSS
- Secure model validation to guard against adversarial attacks
- Customized training to embed AI security best practices
- Penetration Testing (Mobile, Web, AI, Product, IoT, Network & Cloud)
- Secure Software Development Consulting (SSDLC)
- Customized CyberSecurity Services
COE Security helps organizations strengthen AI security through AI system assessments, secure model validation, application and API penetration testing, cloud and network security reviews, data governance, and compliance-focused cybersecurity consulting.
For organizations adopting third-party AI platforms, we can help identify security risks, assess integrations and data flows, evaluate access controls, and develop practical recommendations for safer deployment.
We also support secure software development, vulnerability management, security testing, and customized cybersecurity programs designed around an organization’s operational and regulatory requirements.
Our goal is to help organizations adopt emerging technologies responsibly while protecting sensitive data, reducing cyber risk, strengthening resilience, and supporting compliance.
Follow COE Security on LinkedIn for ongoing insights into AI security, cybersecurity threats, regulatory developments, and safe, compliant AI adoption. Stay informed and cyber safe.
Click to read our LinkedIn feature article