Google AI Chief's Distillation Talk Spurs 20+ Companies to Urge No Premature Rules; Anthropic, OpenAI Ban It

Earlier this year, Google AI chief Jeff Dean discussed AI distillation in a podcast. He said Google discovered the technique while developing AI models, aiming to improve system performance without relying on a single large image recognition model. In February, Dean noted that distillation is a key technology for making smaller models more capable, and that a frontier model must exist first before it can be distilled into a smaller one.

Five months later, distillation has become a hot topic from Silicon Valley to Washington. Distillation refers to using the answers from a chatbot or the output of an advanced AI model to train another model. The practice is controversial because, depending on how it is used, it can allow model developers to create competitive products merely by using the output of companies that have invested millions or billions of dollars in developing training technology.

On Friday, several global technology giants jointly issued an open letter urging policymakers to avoid "premature restrictions" on open-weight AI models. Signatories include more than 20 companies such as Nvidia, Microsoft, Meta, and Palantir. The letter states that distillation is a widely used technique for model improvement, evolution, and verification, where one model's output is used to help train or improve another model.

Aaron Levie, chief executive of Box and a signatory, said that to remain competitive, US companies need access to the best technology wherever it is developed. Levie believes the overall trend is that more innovation leads to more AI progress, which will move toward lower cost and greater efficiency.

Shashi Bellamkonda, research director at Info-Tech Research Group, said many companies incorporate distillation when creating their own models. For example, Nvidia used distillation in training its Llama Nemotron series of models, as detailed in research papers.

Anthropic holds a different view. The company said that preventing illegal distillation is a matter of national security. In a February post, Anthropic stated that it has built systems to prevent state and non-state actors from using AI to develop biological weapons or conduct malicious cyber activities, and that stopping such threats requires rapid, coordinated action among industry participants, policymakers, and the global AI community.

OpenAI and Anthropic prohibit distillation in their terms of service. Bellamkonda said they are essentially implying that unauthorized use of their large models represents potential intellectual property theft. As AI costs soar, companies will do everything possible to improve efficiency.

Pukar Hamal, founder of SecurityPal, which uses AI to automate security assessments, said his company has no objection to using open-weight models. He said this can save significant money, and after completing assessments, the models are hosted on the company's own infrastructure.

A challenge for Anthropic and OpenAI in arguing intellectual property theft is that both companies rely on content from others to build their own models and have been sued for doing so. Max Pritt, a lawyer at Boies Schiller Flexner representing book authors in copyright lawsuits against AI companies, said the government is in a similar position. At least publicly, this administration focuses on protecting tech companies' intellectual property while remaining silent on the unauthorized use of creators' and individuals' intellectual property.