Pros and Cons of Open-Weight Models

Open-weight AI makes model weights public, allowing anyone to download, run, and modify them. The advantage is that users can directly own AI, free from corporate terms of service and restrictions, but at the same time, the disappearance of central gatekeeping makes it difficult to prevent misuse related to hacking, child sexual abuse material, terrorism, and biological weapons.

The current debate focuses on whether open weights should be banned. Proponents argue that since malicious users can obtain models anyway, good-faith users should also be able to use them for defense and security. More than 100 companies, including Microsoft, NVIDIA, OpenAI, Intel, Amazon, Meta, and Hugging Face, recently signed an open letter supporting open-weight AI.

The opposing camp has not emerged clearly. Some Trump administration officials have expressed negative views citing competition with China, but they have not called for a total ban, and Anthropic has taken an ambiguous stance, supporting only open-weight models without dangerous capabilities.

AI safety organizations generally maintain public neutrality. The analysis suggests that while open-weight models are outside central control and difficult to permanently align against human misuse or AI loss of control, it is also difficult to definitively label open weights as uniquely dangerous because even major companies have not yet established methods for AI alignment.

At this point, the gap between closed frontier models and the best open-weight models is only about 6 months. The author believes that open-weight AI may not be sustainable in the long term, but takes the position of waiting for the conclusion to emerge through market and technological development rather than spending political resources on an immediate ban.