Google, following on the heels of OpenAI, revealed a coverage proposal in response to the Trump administration’s name for a nationwide “AI Action Plan.” The tech large endorsed weak copyright restrictions on AI coaching, in addition to “balanced” export controls that “defend nationwide safety whereas enabling U.S. exports and world enterprise operations.”
“The U.S. must pursue an energetic worldwide financial coverage to advocate for American values and help AI innovation internationally,” Google wrote within the doc. “For too lengthy, AI policymaking has paid disproportionate consideration to the dangers, typically ignoring the prices that misguided regulation can have on innovation, nationwide competitiveness, and scientific management — a dynamic that’s starting to shift underneath the brand new Administration.”
One of Google’s extra controversial suggestions pertains to using IP-protected materials.
Google argues that “truthful use and text-and-data mining exceptions” are “important” to AI growth and AI-related scientific innovation. Like OpenAI, the corporate seeks to codify the correct for it and rivals to coach on publicly out there information — together with copyrighted information — largely with out restriction.
“These exceptions enable for using copyrighted, publicly out there materials for AI coaching with out considerably impacting rightsholders,” Google wrote, “and keep away from typically extremely unpredictable, imbalanced, and prolonged negotiations with information holders throughout mannequin growth or scientific experimentation.”
Google, which has reportedly skilled various fashions on public, copyrighted information, is battling lawsuits with information homeowners who accuse the corporate of failing to inform and compensate them earlier than doing so. U.S. courts have but to determine whether or not truthful use doctrine successfully shields AI builders from IP litigation.
In its AI coverage proposal, Google additionally takes situation with sure export controls imposed underneath the Biden administration, which it says “could undermine financial competitiveness objectives” by “imposing disproportionate burdens on U.S. cloud service suppliers.” That contrasts with statements from Google rivals like Microsoft, which in January stated that it was “assured” it might “comply absolutely” with the foundations.
Importantly, the export guidelines, which search to restrict the provision of superior AI chips in disfavored international locations, carve out exemptions for trusted companies in search of massive clusters of chips.
Elsewhere in its proposal, Google requires “long-term, sustained” investments in foundational home R&D, pushing again in opposition to current federal efforts to cut back spending and eradicate grant awards. The firm stated the federal government ought to launch datasets that is likely to be useful for business AI coaching, and allocate funding to “early-market R&D” whereas guaranteeing computing and fashions are “broadly out there” to scientists and establishments.
Pointing to the chaotic regulatory surroundings created by the U.S.’ patchwork of state AI legal guidelines, Google urged the federal government to move federal laws on AI, together with a complete privateness and safety framework. Just over two months into 2025, the variety of pending AI payments within the U.S. has grown to 781, in keeping with a web-based monitoring device.
Google cautions the U.S. authorities in opposition to imposing what it perceives to be onerous obligations round AI techniques, like utilization legal responsibility obligations. In many instances, Google argues, the developer of a mannequin “has little to no visibility or management” over how a mannequin is getting used and thus shouldn’t bear accountability for misuse.
Historically, Google has opposed legal guidelines like California’s defeated SB 1047, which clearly laid out what would represent precautions an AI developer ought to take earlier than releasing a mannequin and by which instances builders is likely to be held chargeable for model-induced harms.
“Even in instances the place a developer supplies a mannequin on to deployers, deployers will typically be greatest positioned to know the dangers of downstream makes use of, implement efficient threat administration, and conduct post-market monitoring and logging,” Google wrote.
Google in its proposal additionally referred to as disclosure necessities like these being contemplated by the EU “overly broad,” and stated the U.S. authorities ought to oppose transparency guidelines that require “divulging commerce secrets and techniques, enable rivals to duplicate merchandise, or compromise nationwide safety by offering a roadmap to adversaries on how you can circumvent protections or jailbreak fashions.”
A rising variety of international locations and states have handed legal guidelines requiring AI builders to disclose extra about how their techniques work. California’s AB 2013 mandates that corporations creating AI techniques publish a high-level abstract of the datasets that they used to coach their techniques. In the EU, to adjust to the AI Act as soon as it comes into power, corporations should provide mannequin deployers with detailed directions on the operation, limitations, and dangers related to the mannequin.
