MIT: the open-model debate pits real benefits against real risks

MIT Technology Review examines the debate over open AI models after an Anthropic report warned that Z.ai's open GLM-5.3 poses a major cyberattack risk. Open weights give researchers and businesses cheaper, local access, but anyone can download and modify them, fueling disputes over safety and concentrated power.

• Anthropic said Z.ai's open GLM-5.3 poses a major cyberattack risk
• All top eight models on the Artificial Analysis leaderboard are closed
• Open models cost less than closed ones and can run on local hardware
• Jensen Huang backs open models; Dario Amodei urges caution