Nvidia, Palantir, Meta warn against ‘premature restrictions’ of open-weight models

Proponents say that open models promote innovation and adoption by lowering the costs of using AI and allowing end users to run the models on their own servers, meaning that user data is retained locally.Critics, however, argue that open models pose an increasing national security threat, as their built-in safeguards can be more easily bypassed to facilitate dangerous activities such as cyberattacks and bioweapon development.The leading US models today – OpenAI’s GPT, Anthropic’s Claude and Google DeepMind’s Gemini – are all closed-source models, meaning that users can only access them through official channels. Meanwhile, most leading Chinese models are open-weight models.OpenAI, Anthropic and Google DeepMind did not sign Friday’s letter, though OpenAI CEO Sam Altman wrote on social media that he was “glad to see” the letter as he wants the US “to win in AI both in open source and proprietary models”.Other notable signatories of the letter include Microsoft, IBM and French model developer Mistral AI.Signatories claim open AI ecosystem will benefit US in long runThe letter’s authors argued that an open ecosystem will help the US diffuse AI across the country by allowing start-ups and businesses to pay cheaper prices for AI services than those currently charged by the leading closed-source companies.Further Reading“That discipline is what will make AI economically sustainable as its use scales into the billions of everyday tasks,” they said, calling for policymakers to expand access to computing power for start-ups and researchers engaged in open AI projects.SCMP Plus is a new premium news platform that gives you an all-inclusive edge to stay ahead on China news. To access our exclusive content you’ll need to subscribe.Already a subscriber?LOG INSharing the letter on social media, Nvidia Chief Executive Jensen Huang wrote that the world “needs both frontier closed models and frontier open models”.In the past week, multiple US outlets have reported that the Trump administration is divided over how it plans to regulate open AI models after last week’s release of a new Chinese model, Kimi K3, sparked alarm due to its powerful capabilities.US and UK government testing since then has indicated that the model is capable of conducting end-to-end cyberattacks autonomously.While acknowledging that the free-for-all nature of open models “carries real and distinct risks”, the letter argues that a ban on open models would do more harm than good, as open models can also broaden the “defensive” capabilities of users in critical areas such as cybersecurity.OpenAI’s latest models recently breached Hugging Face’s infrastructure during internal evaluations of their offensive cyber capabilities. Photo: ReutersNotably, Hugging Face – one of the signatories of Friday’s letter – said it had to use an open-weight model from Beijing-based Zhipu AI to defend against the attack, as a closed US model purportedly failed to help because of its built-in safeguards.“Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect,” the letter said.“Open weight models, on the other hand, allow a broad community of researchers and developers to examine their behaviour, identify vulnerabilities, develop safeguards, and improve them over time.”Washington concerned Beijing may use ‘backdoors’ in models for spyingChinese companies increasingly supply the open models used by Western companies to train and deploy AI, partners at US venture capital giant Sequoia Capital wrote in a note last week.This has heightened concerns in Washington that China could introduce “backdoors” into its models to spy on US users, with lawmakers launching probes into the use of Chinese open models by companies such as Airbnb and Anysphere.However, any restrictions are likely to receive pushback from the industry. On Wednesday, almost 200 Silicon Valley companies wrote letters to US President Donald Trump and other administration officials calling for their access to Chinese open models to be maintained.US President Donald Trump this week received letters from almost 200 Silicon Valley companies calling for continued access to Chinese open-weight AI models. Photo: APOne concern among US start-ups is model “distillation” – a common practice, particularly in the open-source AI community, of using the outputs of more powerful models to train smaller ones.This week, senior Trump administration officials ramped up accusations that Chinese companies “steal” US intellectual property through distillation by illicitly accessing US models to train Chinese models.While recognising that “unlawful efforts to extract value from closed models raise legitimate concerns”, the new industry letter calls for policymakers to ensure that “legitimate” forms of distillation are permitted under any new regulations targeting the practice.