Open models have real benefits. Do they outweigh the risks?

The debate over whether frontier AI models should be open or kept behind closed doors has intensified. Recently, Anthropic released a report arguing that GLM-5.3, a powerful model released by Chinese company Z.ai, poses major cyberattack risks because it is freely available to download and modify. This highlights the ongoing tension between safety advocates, who warn that open models can be misused by bad actors, and proponents of open systems, who argue that open-weight models prevent the mass concentration of power among a few large companies. According to MIT Technology Review, closed models like ChatGPT and Claude are currently more capable and are accessed solely through APIs or apps with safety filters that can block harmful queries and revoke user access. Conversely, open models allow users to run them on local hardware, examine their inner workings and chain of thought, and adapt them for various applications. Academic researchers rely on open models because closed systems restrict them to observing only inputs and responses. Businesses also benefit from open models by processing sensitive data securely on local hardware and generally incurring lower costs. However, experts note that open models are not without commercial drawbacks; for instance, some companies may use open releases as an extractive approach, integrating community improvements without compensating outside contributors. Ultimately, deciding whether to support open models involves complex trade-offs between safety, national security, and democratization.