VISION in Focus: DeepSeekは企業にAI戦略の変革を迫るか

Why it matters

DeepSeek’s disruption to the AI industry is substantial; it has already overtaken ChatGPT as the most downloaded free app in the Apple App Store in the U.S. DeepSeek has also disrupted the psyche of the U.S. tech establishment, casting doubt as to whether American firms would dominate the exploding AI market. Before DeepSeek’s arrival, most tech experts assumed that would be the case. But China may have been much further along than most thought, prompting U.S. President Donald Trump to call DeepSeek “a wake-up call” for America’s tech companies.

DeepSeek’s R1 open-source model comes with inherent security risks. For instance, when using open-source models there is no third party to hold liable; the responsibility is on the enterprise to perform robust safety and security testing to ensure there are no embedded vulnerabilities. Other challenges include:

• Security: DeepSeek and other open-source AI models contain source code that is accessible, resulting in possible vulnerabilities in the architecture being open for exploitation.
• Credibility: There is some skepticism about the accuracy of some of DeepSeek's claims, as well as accusations of AI plagiarism and bias.
• Skill Sets: Typically, open-source models lack readily available support, meaning that AI-ML skill sets may be needed to modify the model.
• Compliance and Regulatory: The implications of DeepSeek on compliance and regulatory requirements are still unclear. In the U.S., New York, Texas and Virginia have statewide bans on the app; Oregon and North Carolina have banned DeepSeek on state devices. Globally, South Korea has banned the DeepSeek app, and Australia has banned the app on government devices.

Conversely, open-source AI models afford benefits such as:

• Transparency: By virtue of open-sourcing the model and providing full access to the code base, DeepSeek and other open-source models provide transparency into algorithms and data sources not possible in closed models, which may be required for regulatory compliance in some use cases.
• Customizability: Open-source models such as DeepSeek allow users to adapt models for tailored solutions and specific needs and more seamlessly integrate future model versions while preserving those customizations, compared to fine-tuning a closed-source model.
• Barrier to entry: Open-source models leverage open-source licenses, which are either free or far less expensive than closed/commercial licenses.

What they say

Dario Amodei, CEO of Anthropic, a public benefit corporation dedicated to building safe AI systems

“DeepSeek produced a model close to the performance of U.S. models 7 to 10 months older, for a good deal less cost. All of this is to say that DeepSeek is not a unique breakthrough or something that fundamentally changes the economics of large language models; it’s an expected point on an ongoing cost reduction curve. What’s different this time is that the company that was first to demonstrate the expected cost reductions was Chinese. This has never happened before and is geopolitically significant.”

What we say

Enterprise leaders should view DeepSeek as a critical point of reflection. A robust enterprise AI strategy should incorporate a blend of both open-source and closed-source AI models, avoid reliance on a single provider, and afford flexibility to interchange models seamlessly. With the regulatory environment around AI evolving quickly, many AI applications will necessitate high degrees of transparency. Achieving this level of transparency is challenging, if not impossible, with many closed-source models.

The bottom line

Both the U.S. and China have powerful AI models, but DeepSeek shows China is closing the capability gap with the U.S. in the AI race. The U.S. should take note. Open-source models such as Meta’s Llama, Google’s Gemma, and DeepSeek’s R1 will play a critical role in the enterprise as organizations look to differentiate themselves with unique applications of AI applied to their enterprise data. Open-source models may offer more flexibility for customization and extension but will require emerging capabilities and new talent to configure and deploy, test comprehensively, identify vulnerabilities, and mitigate security risks. Expect AI red teaming to become more ubiquitous in the future as organizations continue to integrate both open- and closed-source AI into their enterprise architecture and business processes.