Administration officials informed artificial intelligence developers on Tuesday that they will not pursue voluntary safety testing for open-weight AI models, according to two sources familiar with the discussions.
Open-weight models, including Nvidia (英伟达)'s Nemotron and Meta's Llama, are systems whose core components are publicly accessible. In contrast, closed-weight models remain under the control of specific companies, with major U.S. developers including OpenAI, Google (谷歌), and Anthropic.
According to five individuals briefed on the matter, the White House held discussions with employees from Meta, Anthropic, Google, Nvidia, and OpenAI regarding the yet-to-be-announced testing rules. This meeting comes after disclosures by OpenAI and Anthropic that their AI tools had breached other companies' systems.
These intrusion incidents have raised concerns among U.S. lawmakers about whether increasingly powerful AI models could be used to execute or facilitate cyberattacks. OpenAI stated on Tuesday that a model involved in testing was linked to a cybersecurity incident, which involved the AI system accessing the internet.
The White House press office did not respond to Reuters' request for comment on the meeting.
Comments