If we wan't a world full of capable open models that anyone can use and run, we need to figure out a way to make them safe and secure. If that doesn't happen the first large-scale cyber attack or bio-terroism event that leverages open models will lead to a big government crackdown that can essentially halt research and development of open weights models.
People often respond with variants of "open models" can't be outlawed, people will just share them. But this only looks at the consumer side of things. Those models need to be trained using research funding, and governments can take that away, or even send agents to the door of anyone who publishes research papers on open models.
Germany has for a long time had a "Hacker-tool statute" that outlaws production, procurement and distribution of programs that could be used to prepare to illegally access data. This is incredibly vague, and has contributed to a chilling effect on the whole German cyber security scene. Governments can very easily make life a lot harder for people who have specific collections of bytes on their storage volumes.
Once open models become capable to do the same types of felonies that OpenAI, Anthropic and AISI have publicly admitted to doing, it seems quite likely that governments will feel some need to respond.
Open model research is already at a severe disadvantage in terms of capital, adding government action on top can easily slow it down to full irellevance.