Open AI models: benefits weighed against risks
The debate over open models is more complicated than it seems.
Policy#04
Policy
Image: MIT Technology Review
The short version
- Anthropic released a report arguing that GLM-5.3 poses a major cyberattack risk.
- The model was released in August by the Chinese company Z.ai and is freely available to download and modify.
- As of publication, the top eight models on the Artificial Analysis LLM leaderboard are all closed, though open models are only a handful of months behind.
- Open models let researchers study their inner workings, and businesses can run them on their own hardware to keep sensitive data secure.
- Once someone downloads an open model, there is no filter checking their prompts and no way to revoke their access.
Why it matters
Open models sit at the center of the debate over who controls frontier AI. Supporters argue they spread power beyond a few companies, while critics warn they are easier to misuse. Any policy involves trade-offs between safety and democratization.
“The clearest, biggest argument [for open models] is around preventing mass concentration of power—it is about making sure that this foundational technology is a public good,”