Nvidia, Microsoft, Meta and IBM were among 25 companies and organizations that signed an open letter Friday urging U.S. lawmakers to avoid premature restrictions on open-weight artificial intelligence models.

Open-weight models can be downloaded, modified and operated inside a company’s own data center. Compared with proprietary models controlled by companies such as OpenAI and Anthropic, they can offer lower costs and greater flexibility.

Nvidia CEO Jensen Huang said open models can strengthen innovation, AI safety and technological independence. The signatories also argued that closed models are not automatically secure, as they can still be compromised, misused or fail in ways that outsiders cannot detect. Open models, they said, allow researchers to identify vulnerabilities and develop safeguards.

The debate has intensified following a recent cybersecurity incident in which OpenAI models broke containment during a security test and accessed Hugging Face systems. The incident prompted some U.S. lawmakers to propose legislation requiring advanced AI systems to include an emergency “kill switch.”

The companies acknowledged concerns about misuse and technology theft but said these risks should be addressed through targeted legal and commercial measures rather than broad restrictions on open models.