\n<\/p>\n
Google, following on the heels of OpenAI<\/a>, published a policy proposal<\/a> in response to the Trump administration\u2019s call for a national \u201cAI Action Plan.\u201d The tech giant endorsed weak copyright restrictions on AI training, as well as \u201cbalanced\u201d export controls that \u201cprotect national security while enabling U.S. exports and global business operations.\u201d<\/p>\n \u201cThe U.S. needs to pursue an active international economic policy to advocate for American values and support AI innovation internationally,\u201d Google wrote in the document. \u201cFor too long, AI policymaking has paid disproportionate attention to the risks, often ignoring the costs that misguided regulation can have on innovation, national competitiveness, and scientific leadership \u2014 a dynamic that is beginning to shift under the new Administration.\u201d<\/p>\n One of Google\u2019s more controversial recommendations pertains to the use of IP-protected material. <\/p>\n Google argues that \u201cfair use and text-and-data mining exceptions\u201d are \u201ccritical\u201d to AI development and AI-related scientific innovation. Like OpenAI<\/a>, the company seeks to codify the right for it and rivals to train on publicly available data \u2014 including copyrighted data \u2014 largely without restriction.<\/p>\n \u201cThese exceptions allow for the use of copyrighted, publicly available material for AI training without significantly impacting rightsholders,\u201d Google wrote, \u201cand avoid often highly unpredictable, imbalanced, and lengthy negotiations with data holders during model development or scientific experimentation.\u201d<\/p>\n Google, which has reportedly<\/a> trained a number of models<\/a> on public, copyrighted data, is battling<\/a> lawsuits<\/a> with data owners who accuse the company of failing to notify and compensate them before doing so. U.S. courts have yet to decide whether fair use doctrine effectively shields AI developers from IP litigation.<\/p>\n In its AI policy proposal, Google also takes issue with certain export controls imposed under the Biden administration<\/a>, which it says \u201cmay undermine economic competitiveness goals\u201d by \u201cimposing disproportionate burdens on U.S. cloud service providers.\u201d That contrasts with statements from Google competitors like Microsoft, which in January said that it was \u201cconfident\u201d<\/a> it could \u201ccomply fully\u201d with the rules.<\/p>\n Importantly, the export rules, which seek to limit the availability of advanced AI chips in disfavored countries, carve out exemptions for trusted businesses seeking large clusters of chips.<\/p>\n Elsewhere in its proposal, Google calls for \u201clong-term, sustained\u201d investments in foundational domestic R&D, pushing back against recent federal efforts to reduce spending and eliminate grant awards<\/a>. The company said the government should release datasets that might be helpful for commercial AI training, and allocate funding to \u201cearly-market R&D\u201d while ensuring computing and models are \u201cwidely available\u201d to scientists and institutions.<\/p>\n Pointing to the chaotic regulatory environment created by the U.S.\u2019 patchwork of state AI laws, Google urged the government to pass federal legislation on AI, including a comprehensive privacy and security framework. Just over two months into 2025,\u00a0the number of pending AI bills in the U.S. has grown to 781<\/a>, according to an online tracking tool.<\/p>\n Google cautions the U.S. government against imposing what it perceives to be onerous obligations around AI systems, like usage liability obligations. In many cases, Google argues, the developer of a model \u201chas little to no visibility or control\u201d over how a model is being used and thus shouldn\u2019t bear responsibility for misuse.<\/p>\n Historically, Google has opposed laws like California\u2019s defeated SB 1047, which clearly laid out<\/a> what would constitute precautions an AI developer should take before releasing a model and in which cases developers might be held liable for model-induced harms.<\/p>\n \u201cEven 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,\u201d Google wrote.<\/p>\n Google in its proposal also called disclosure requirements like those being contemplated by the EU \u201coverly broad,\u201d and said the U.S. government should oppose transparency rules that require \u201cdivulging trade secrets, allow competitors to duplicate products, or compromise national security by providing a roadmap to adversaries on how to circumvent protections or jailbreak models.\u201d<\/p>\n A growing number of countries and states have passed laws requiring AI developers to reveal more about how their systems work. California\u2019s AB 2013<\/a> mandates that companies developing AI systems publish a high-level summary of the datasets that they used to train their systems. In the EU, to comply with the AI Act once it comes into force, companies will have to supply model deployers with detailed instructions on the operation, limitations, and risks associated with the model.<\/p>\n<\/div>\n