Google calls for loosening copyright for AI training

Google has released a policy proposal in response to the Trump administration’s request for a national "AI Action Plan," advocating for loosening copyrights on AI training and more flexible export restrictions to preserve the U.S.’s technological leadership. The company’s proposal largely echoes OpenAI’s position, but goes further in its push to give AI companies unrestricted access to copyrighted materials for training models, TechCrunch reports.

One of the most controversial parts of Google’s proposal concerns intellectual property and AI development. The company argues that "fair use" and exceptions for text and data analysis are "critically important" to scientific progress and innovation in AI. Like OpenAI, Google wants to enshrine the legal right to train AI models on publicly available data, including copyrighted content, without obtaining direct permission from copyright holders or paying compensation.

"These exceptions allow for the use of copyrighted, publicly available material for AI training without significantly impacting rightsholders," Google said. The company also notes that without such exceptions, AI developers would be forced to engage in long, complex, and unpredictable negotiations over data use, which could slow progress in AI.

Google is already facing numerous lawsuits from copyright holders who accuse the company of illegally using their materials to train AI. U.S. courts have yet to determine whether the "fair use" doctrine applies to AI developers.

In addition to copyright, Google is also pushing for easing export restrictions imposed by the Biden administration. The company says current rules that limit access to high-performance AI chips in some countries could undermine U.S. economic competitiveness and negatively impact American cloud services.

This position differs from Microsoft's approach, which earlier this year said it was "fully confident" in its ability to comply with export regulations.

Current export restrictions provide exemptions for trusted American companies that use large computing clusters of AI chips, but Google insists that further easing of these rules will better balance national security and AI development.

Google also notes the fragmentation of AI regulation in the US, with 781 AI-related bills currently pending in various states. The company is calling on the federal government to enact a single national AI law that would set clear privacy and security standards and simplify compliance for companies.

Google also opposes laws that hold AI developers liable, such as California's rejected bill SB 1047, which would have held companies liable for harm caused by their models.

"AI developers typically has little to no visibility or control, and thus shouldn’t bear responsibility for misuse," Google says. "Even in cases where a developer provides a model directly to deployers, deployers will often be best placed to understand the risks of downstream uses, implement effective risk management, and conduct post-market monitoring and logging," the document says.