AI News Roundup – Nvidia wins approval to sell advanced AI chips in China, U.S. government expands AI usage, AI training dataset found to contain personally identifiable information, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• • The New York Times reports on successful efforts by Nvidia to win U.S. government approval to sell some models of AI chips in China. This past April, the Trump administration shut down sales from Nvidia to the country, citing national security concerns. Nvidia even went as far as to design an AI chip specifically for the Chinese market to appease regulators, though this particular processor, the H20, is less advanced than those sold by Nvidia elsewhere in the world. The company’s efforts, often involving behind-the-scenes lobbying by allies, culminated in a visit to The White House by Nvidia CEO Jensen Huang, soon after which the U.S. government indicated that it would grant export licenses to Nvidia for selling the H20 chip in China. Sources told the NYT that Huang “argued that American chips should be the global standard and that the United States was making a grave mistake by ceding the giant Chinese market to homegrown rivals,” and President Trump eventually gave his direction to lift the export restrictions. 
• • The Washington Post reports on the U.S. federal government’s recent push to expand the usage of AI technologies at many levels of government work. While Elon Musk has departed the administration, his signature DOGE project has inspired many projects at agencies throughout the federal government aimed at shrinking the federal workforce. Several AI projects were mentioned: over 25,000 personnel within the Department of Defense use an AI tool known as NGA Maven, which processes satellite and drone imagery to identify possible targets. The tool is being expanded to also process audio and text information. Additionally, the Federal Aviation Administration is exploring AI software to aid the work of air traffic controllers, especially in automating repetitive tasks and analyzing air crash data. Further, the U.S. Patent and Trademark Office is considering ways in which AI can replace tasks performed by patent examiners, including AI search tools for searching for similar patents or patent documents, while the Transportation Security Administration is continuing to roll out facial recognition at airport security checkpoints. Further AI initiatives in the federal government are expected to be tested in the coming months. 
• • The MIT Technology Review reports that one of the largest AI training datasets was found to contain millions of items containing personally identifiable information. According to a new study, the major AI training dataset known as DataComp CommonPool contains millions of images of passports, credit cards, birth certificates, and other documents containing sensitive personal information. According to the researchers, however, who only audited 0.1% of the data within the dataset, the true number of images containing such information could be in the hundreds of millions. The study calls for changes to web scraping policies by AI companies and for stronger privacy laws to protect consumers. 
• • National Public Radio reports on a new study of U.S. teenagers’ AI usage. The study, conducted by the nonprofit Common Sense Media, found that 52% of the teenagers use AI “companions” (e.g., AI systems that act as friends for socialization or other chatting) at least a few times a month, primarily for “entertainment purposes.” However, only a third of the teenagers surveyed described these AI chats as the same or more satisfying than talking with human beings, and a similar percentage reported being disturbed by something the AI companion said during a chat. The nonprofit recommends that no one under the age of 18 use AI companions at all, as socialization is important for developing minds, and that AI companions have been shown to lead to addictive behaviors. 
• • Fortune reports on a pilot program from Delta Air Lines that will use AI-powered technology to create individualized pricing for air tickets. The program, currently used for 3% of the fares offered by the company, takes in traveler information such as method and date of booking as well as other personal information and feeds it into an AI system to calculate a price that could vary from person to person. Delta President Glen Hauenstein told investors that the pilot program has shown “amazingly favorable unit revenues.” Consumer privacy advocates were quick to criticize the practice, with one activist telling Fortune that Delta is “trying to see into people’s heads to see how much they’re willing to pay,” though a Delta spokesperson said that the airline has “zero tolerance for discrimination.” The company is planning for 20% of its fares to be AI-determined by the end of 2025.