硅谷上演“开源保卫战”：OpenAI、Anthropic被曝游说封堵开源模型

Odaily reported that multiple informed sources have revealed that OpenAI and Anthropic have engaged in closed-door communications with U.S. regulatory authorities in Washington, aiming to push for stricter restrictions on open-source models. What was once a debate over technical approaches has now escalated into a battle over the right to set industry rules. Open-source models have lowered the barrier for businesses and developers to use advanced AI technology, and have also helped Chinese models quickly gain a global user base. Overseas developers and companies can directly download, modify, and deploy these models without incurring high API call costs.

This has also put competitive pressure on OpenAI and Anthropic. The two companies are conveying to U.S. regulators their concerns about the development of Chinese open-source models, hoping that the Trump administration will strengthen restrictions on them.

At the same time, they continue to emphasize the potential security risks posed by advanced AI models. The two companies argue that as model capabilities continue to approach or even surpass human levels, if model weights are fully open-sourced, any organization can download, modify, and deploy them, which would increase the risk of model misuse.

OpenAI has previously stated publicly that it hopes to establish a government-led AI safety assessment system, implementing mandatory safety tests for some advanced models. Anthropic has repeatedly stressed that open-source models could lower the threshold for the proliferation of dangerous capabilities, and therefore, stricter management measures should be adopted. (Science and Technology Board Daily)