What are Open Source AI Models?

As the name suggests, an open sourced AI model allows for anyone to freely edit, modify, and distribute an AI model. The point of making an AI model is to invite collaborators to tackle the challenges behind creating an AI model. Due to the overwhelming skill and technique needed to code a simple AI model let alone a complex LLM, AI researchers are always open to team up with other researchers in order to create ethical and viable AI for human welfare. 

China has embraced open-source AI by investing in projects such as DeepSeek. Known for producing similar results like OpenAI and Claude at half the expense, DeepSeek has been trending in the AI world from its release in 2023 and its improved models in recent months. In China, DeepSeek is much more than a research tool—it is part of a much bigger national effort to accelerate its domestic AI development. DeepSeek is regularly updated and widely distributed while being supported by private and state-backed initiatives. 

Yet as projects like DeepSeek gain global traction, they also highlight a deeper question at the heart of open-source AI: how do we balance innovation with accountability? The advantages of open-source are huge—more researchers working on a model will inevitably lead to bigger LLMs and stronger AI. However, it can be easy to get lost in the benefits and forget about the security risks and complexity of the model itself.

What are the Security Risks that come with Open Source?

Depending on whether the AI model depends on public or open-source data, data breaches and privacy leaks can be a huge problem. Since the accessibility of the model is not privatized, anyone can go into the code and maliciously edit it. Even worse, the data breach usually cannot be traced back to the original attacker.

As proof of this, Meta’s Llama was leaked online in February 2023, where in its initial release the model’s files were circulated on online forums including the software piracy files. Although nothing happened to the model due to the cancellation of the model, the fact that files were so easily leaked caused public concern of whether or not we should continue to have open sourced AI models.

This tension hasn’t slowed the momentum—particularly in China, where open-source AI is not only thriving but strategically embraced by tech giants like Huawei. Just a few weeks ago, Huawei released 2 additional models for open source. Hoping that more people will feel more incentivized to use other Huawei products such as its Ascend AI chips, their open source models aim to build more traction for the company as a whole. 

Instead of learning from the mistakes of past companies, China is sprinting ahead with the hopes of not repeating history and reaping the benefits of the AI market. However, without actual security clearances and regulation, history is bound to repeat itself. 

How should we move forward?

Should openness be sacrificed for safety? Who gets to decide what constitutes “harm” or “misuse”? These types of questions should be asked of companies who use open-source AI in order to make sure that governance is not an afterthought. 

Possible governance models aren’t too complicated as well. Implementing gated access using licenses and vetting processes can help combat misuse of open access. Some famous models include Do-Ocracy and Founder-Leader governance models, which highlight how decisions should be made and who should be in charge of critiquing and reviewing the AI. However, these models do not look at the impact that open-source models have on the technological progress of smaller countries. With more and more restrictions on what AI can provide, countries that rely heavily on open source models will suffer from these governance models. Maintaining a key balance will be imperative in order to reap the benefits of open-source AI models.

The time to create these models is now—already we are seeing regulatory frameworks like the EU AI Act starting to be implemented within the European Union. As open-source AI continues to evolve, so must our approach to governance. We need frameworks that encourage collaboration while safeguarding against misuse. Striking that balance will define the future of responsible innovation.