OpenAI and Anthropic Find Common Ground: Unite Against Open-Weight AI Models

Anthropic CEO Dario Amodei is one of the main advocates of this position. He has publicly stated multiple times that open-weight models are harder to keep safe because once weights are released, developers lose control over the model. Specifically, Amodei pointed out three key issues: first, inability to revoke access for malicious actors; second, inability to update safety guardrails and filtering mechanisms; third, inability to prevent the model from being used for harmful purposes. He believes that for frontier-level AI models, developers should retain some control capability to ensure timely response to newly discovered security threats. This view was echoed by OpenAI, which stated in its latest safety blog: 'Model safety and safeguards must keep pace with rapidly advancing capabilities.'

However, opposition is equally strong. Open-source advocates argue that open-weight models are actually safer because transparency allows independent security researchers to review code, discover vulnerabilities, and propose fixes. They point out that closed models are merely 'security through obscurity'—once internal safety measures are breached, external users have no protection. Open-weight models, on the other hand, allow the community to collectively build defense mechanisms. Additionally, open weights promote innovation and competition, preventing a few large companies from monopolizing AI technology. Companies like Meta and Mistral continue to insist on the open path, believing it is key to AI democratization.

The background of this debate is that Chinese AI companies are aggressively pursuing open-weight strategies. Moonshot AI's Kimi K3, Alibaba's Qwen series, and other Chinese AI labs are all actively releasing open-weight models. These models have approached or even surpassed US top models in performance in some aspects. OpenAI and Anthropic's opposition stance has therefore been interpreted by some as a strategic move targeting China—using regulatory restrictions to block the spread of Chinese open-weight models. While this interpretation is oversimplified, it does reflect the geopolitical dimension in AI competition. Trump administration adviser David Sacks explicitly warned that regulation in the name of safety could become 'regulatory capture'—rules intended to improve AI safety instead entrench large companies by making it harder for competitors to release models.

From a practical impact perspective, this debate could profoundly change the AI industry's regulatory landscape. If governments impose stricter open-weight restrictions on frontier AI models, several consequences may follow: first, small AI companies and research institutions will have more difficulty accessing advanced models for research and innovation; second, AI technology development may become more concentrated in the hands of a few large companies; third, the power of the open-source community may be weakened; fourth, US-China AI competition may further intensify. However, if no restrictions are placed on open weights, there is indeed a risk of model misuse, especially given rapidly advancing AI capabilities. Finding a balance between safety and openness will be the core challenge for AI policy-making in the coming years.