Global AI Governance Under Fire: The DeepSeek Problem Explained

What happens when one AI giant accuses another of stealing its secrets? The OpenAI vs. DeepSeek dispute might just redefine the rules of AI development.

The creator of ChatGPT, OpenAI, has accused DeepSeek, a rising Chinese AI startup, of using a forbidden method to train its models, potentially violating OpenAI’s terms of use. This allegation has sparked a broader debate about intellectual property rights, data privacy, and the future of AI governance.

Adding fuel to the fire, DeepSeek is now facing regulatory scrutiny from governments worldwide...

In this blog post, we’ll explore the OpenAI-DeepSeek controversy, the regulatory challenges DeepSeek faces, and what this means for the future of AI governance.

The OpenAI vs. DeepSeek Controversy

At the center of OpenAI’s allegations against DeepSeek lies a machine learning technique known as knowledge distillation.

OpenAI’s suspicion about DeepSeek

According to their Terms of Use, OpenAI (creator of ChatGPT) prohibits the training of a new AI model by using another larger, pre-trained model, a practice referred to as knowledge distillation.

OpenAI policies explicitly prohibit:

• Reverse engineering, decompiling, or discovering underlying components of its models.
• Using OpenAI’s output to develop competing models.
• Automatically extracting data from OpenAI’s services.

If OpenAI can prove that DeepSeek trained its AI using distilled knowledge from OpenAI’s models, this would constitute a clear violation of these rules - potentially opening the lawsuit door.

In simpler terms, knowledge distillation enables AI developers to take an advanced model, extract its knowledge, and use it to train a more efficient version. This method helps reduce computational costs while maintaining high performance.

However, proving knowledge distillation is pretty hard to do. Unlike traditional intellectual property violations, AI model outputs are not at all easy to trace like copyrighted text, images, or code. DeepSeek could argue that its models were trained using publicly available data or that its training methods were developed on their own.

OpenAI’s Position on the Issue

‍Although OpenAI hasn’t filed a lawsuit yet, it has made strong statements about AI knowledge theft:

OpenAI Spokesperson:

“Our models and services are protected by strict Terms of Use. We take violations seriously and will pursue legal action when necessary to protect our intellectual property and ensure fair competition in the AI ecosystem.”

‍

Opinion of AI experts

‍Leading AI experts have raised broader concerns about knowledge distillation and the challenges of proving it.

Dr. Fei-Fei Li, AI Researcher & Professor at Stanford University:

“Knowledge distillation is a widely used technique in AI, but this case raises an important legal question - can companies claim ownership over model behaviors and outputs? If OpenAI succeeds, it could change how AI models are trained worldwide.”

‍

Ben Goertzel, CEO of SingularityNET:

The real issue is whether AI outputs can be considered proprietary. If OpenAI wins this battle, it could lead to more restrictive AI development policies, limiting innovation.”

‍

Why it’s important

The dispute raises questions far beyond a corporate dispute between OpenAI and DeepSeek. If OpenAI decides to pursue legal action - it could set a groundbreaking precedent in AI governance and intellectual property rights as regulators themselves are figuring it out as they go.

The world is no stranger to some AI laws and privacy restrictions which we will be discussing later on as well, but a case of this magnitude is sure to set some ground rules in place as well. We have our eyes opened on where this goes.

Global Regulation: DeepSeek under fire

The Italian DPA’s Inquiry into DeepSeek

The Italian Data Protection Authority (DPA) has started an inquiry into DeepSeek, citing significant concerns about the company’s data practices and compliance with European Union regulations. In a formal request, the DPA demanded detailed information from Hangzhou DeepSeek Artificial Intelligence and its Beijing-based affiliate, the entities behind the DeepSeek chatbot service.

The DPA has requested info on:

• What personal data does DeepSeek collect?
• Where does the data come from?
• For what purposes is the data used?
• What is the legal basis for this data collection?
• Are user data and AI training datasets stored on servers located in China?

DeepSeek now has 20 days to respond to the inquiry (until February 17th), and failure to comply could result in significant penalties or even a ban on its services in Italy. Its mobile app no longer appears in both the Google and Apple App Stores in Italy.

Italy’s data privacy regulators are no strangers to enforcing against AI companies. It briefly banned ChatGPT over alleged privacy violations when the app caught global attention in 2023. In December 2024, it fined its parent company OpenAI €15 million.

Australia’s Ban on DeepSeek

In a bold move, the Australian government banned DeepSeek from all government devices and systems, citing national security concerns. The ban, which applies to both the web platform and mobile app, requires government entities to remove any existing DeepSeek installations and prevent further use of the service.

The Australian government says the decision is not based on DeepSeek’s Chinese origins but rather on the “unacceptable risk” the AI startup poses to national security.

This move aligns with a broader trend of governments restricting the use of foreign technology in sensitive areas, particularly when it involves AI and data processing (like the Italy case above).The Australian government's ban includes:

• Blocking the installation or use of DeepSeek on all government-issued devices.
• Removing any existing DeepSeek software from government systems.
• Preventing public sector employees from using DeepSeek’s services for official purposes.

The ban relates to some previous restrictions placed on TikTok and Huawei, both of which were scrutinized over their ties to China.

While private citizens in Australia can still use DeepSeek, the government is sending a clear message: foreign AI services must first prove they are secure.

Will DeepSeek Survive in the EU?

All this regulatory trouble in Italy and Australia raises a critical question: Can the company maintain a presence in Western markets?

Well, there are two possible scenarios:

1. DeepSeek adapts to regulations

If DeepSeek adjusts its data policies to comply with GDPR and other international standards, it might still have a chance to remain active within the EU.

However, this would require some changes on their side. Most important - transparency about data collection and processing methods, which could possibly slow down its AI development and market penetration. The company may need to invest heavily in legal and compliance teams, modify its data practices, and set local data storage facilities to ease regulators.

2. DeepSeek ignores regulations

If DeepSeek fails to comply or if authorities deem it a national security risk… more governments could follow Australia’s lead in banning its products (even if partially). This could significantly limit DeepSeek’s ability to expand internationally, forcing it to focus primarily on the Chinese market and non-Western regions.

Regardless of which path DeepSeek takes, its regulatory battles are a sign of what’s to come in AI governance - “deep into the unknown” as its slogan states.

DeepSeek in the US - Trump’s Executive Order on AI

In a move to reshape U.S. AI policy, President Donald Trump signed an executive order which supports American leadership in the AI field. The order, which revoked certain AI policies enacted under the Biden administration, outlines a vision for AI development that prioritizes innovation, economic competitiveness (aka funding), and national security.

This policy shift raises critical questions about how the U.S. government will handle foreign AI companies like DeepSeek.

Key Points of the Executive Order

Trump’s executive order revokes some AI regulations from the Biden administration and replaces them with a more pro-business AI strategy. Because it wants to boost U.S. AI innovation - it may lead to direct consequences for foreign AI companies like DeepSeek trying to enter the American market. Lets explore below:

This means U.S.-based AI companies like OpenAI, Anthropic, and Google DeepMind are in good soil to thrive. They are expected to receive preferential treatment through government funding, regulatory relief, and national security protections. None of which of course apply to DeepSeek.

Conclusion

As AI companies compete to dominate the industry, governments and regulators are struggling to keep up, crafting new policies on data privacy and security risks. DeepSeek’s future in Western markets remains uncertain. If it complies with regulations, it may slow its AI development but retain access to global markets. If it refuses, more governments could follow Australia’s lead, limiting DeepSeek’s influence outside China.

Beyond DeepSeek, this case raises critical questions:

• Who owns AI-generated knowledge?
• How do we regulate AI training without hurting innovation?
• Where does copyright even stand?

We are covering these topics in our next blog for AI governance, which you can read here.

It is an immense topic and one we are following excitedly. Hope you are too!