Open Models Have Real Benefits. Do They Outweigh the Risks?
MIT Technology Review, Tuesday, October 13th, 2026
An explainer on the open versus closed AI model debate, weighing democratization and research benefits against misuse risks.
Prompted by Anthropic's report that Z.ai's open GLM-5.3 poses cyberattack risk, the explainer lays out the debate over open-weight models.
It defines open versus closed models and notes that the top eight models on the Artificial Analysis leaderboard are closed, with open models a few months behind.
Open models can be run locally, retrained and inspected, which researchers need in order to study model internals and chains of thought.
The piece also describes commercial incentives on both sides and the trade-offs between safety and democratization that any policy must weigh.