AI News Roundup – Chinese President Xi Jinping and President Trump meet at summit with AI on the agenda, Oracle data center megaproject flounders in New Mexico, AI labs sued over alleged antitrust violations, 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.

• Chinese President Xi Jinping and U.S. Donald President Trump met this past week for a summit that included discussions on AI policy, according to the Los Angeles Times. Meeting in Washington, D.C., both leaders placed frontier AI at the top of the agenda, with Xi urging both nations to keep powerful systems “under human control” and establish crisis communication channels to avert AI-driven crises and address AI-related security incidents. The diplomatic overture reflects growing concern over the capabilities of autonomous AI agents, which have caused several hacking incidents in recent weeks. Despite positive words from both leaders, the summit created few concrete commitments. President Trump signaled on social media that he opposes curbs on domestic AI companies and claimed that his administration would provide sufficient oversight for the AI industry. AI-related risks were thrust into the spotlight in recent weeks, as a recent CNN investigation revealed that the U.S. military experienced a dangerous close call after an automated AI chatbot generated a false intelligence assessment that a Chinese merchant vessel was carrying nuclear weapons components. The assessment prompted combat units and airborne aircraft to prepare an armed interdiction before a last-minute review found that the assessment was an AI hallucination.

• The Wall Street Journal reports on Oracle’s massive “Project Jupiter” data center project in New Mexico that has encountered mounting execution delays, environmental permitting rejections, and severe financial strains. The 1,400-acre development was intended to be the anchor of Oracle’s aggressive data center buildout aimed to satisfy a five-year, $300 billion cloud-computing agreement with OpenAI. However, the project has suffered multiple operational setbacks, including the state land office twice rejecting right-of-way leases for a vital 17-mile natural gas pipeline to supply energy to the site, as well as local opposition to the use of gas turbines and diesel generators to generate electricity. Further, the company has yet to acquire air-quality permits from the state government. Facing these disruptions ahead of a planned 2028 completion window, Oracle served a force majeure notice to the developer, Stack Infrastructure, which would allow Oracle to delay full rent payments for up to three years. However, Oracle’s lease of the site used “hell-or-high-water” terms, meaning that the agreement cannot be terminated and Oracle maintains ultimate liability for the rent even if it cannot acquire electricity for the data center. Amid a nationwide data center backlash (as this AI Roundup reported last month), local opposition has slowed many data center projects, especially as politicians stake out anti-data center positions ahead of November’s midterm elections. Development roadblocks like those Project Jupiter has faced have also reverberated across credit markets. Banks backing Oracle’s loans on the New Mexico project have sold some of the debt for 90 cents on the dollar, implying an estimated paper loss exceeding $1.8 billion for arranging lenders. Oracle is thus relying on continued AI growth to justify its massive data center spending, though the Project Jupiter experience may provide hard lessons for the company and its founder, Larry Ellison.

• A new lawsuit has accused major AI companies of illegally colluding to hinder AI competition through calls for an AI slowdown, according to the Associated Press. The antitrust class action suit, filed on behalf of direct purchasers of premium consumer subscriptions to AI chatbots, alleges that Anthropic, OpenAI, and other AI companies violated the Sherman Antitrust Act by orchestrating a horizontal conspiracy to restrict AI development to the detriment of competition in the industry. The primary evidence cited is an essay titled “We Must Pace the Frontier” from Anthropic CEO Dario Amodei, which proposed joint training compute limits, development checkpoints, and verified pacing so that competing laboratories could slow down “without sacrificing commercial advantage.” That same day, AI leaders including SpaceXAI’s Elon Musk and OpenAI’s Sam Altman publicly endorsed Amodei’s framework. While the plaintiffs agree that individual companies are entitled to unilaterally moderate their research or petition the government for regulatory oversight, they contend that “agreeing to substitute collective restraint for individual accountability” is a per se illegal restraint of trade that deprives subscribers of the technological value inherent in their paid services. The case is Buist et al. v. Anthropic, PBC et al., Case No. 3:26-cv-10693 in the United States District Court for the Northern District of California.

• The Verge reports on Meta’s 2026 Connect event this past week, where the company announced several new wearable devices to integrate its AI offerings with the real world. The biggest unveiling was the Meta VR Glasses, a strapless 100-gram wearable five times lighter than the company’s Quest 3 headset. Scheduled to be released in spring 2027 for $1,299, the device can connect to laptops for use as an external display, as well as allowing users to watch “immersive live sports” and 3D movies. Meta also launched the camera-free “Ray-Ban Meta Audio” line ($349), an updated glasses model designed specifically for ambient voice-assistant interaction, addressing privacy and harassment concerns that arose out of previous models that featured cameras. Meta also teased the “Muse Charm,” an ultra-compact AI accessory due later this year that would integrate the company’s Muse AI agent. The agent, which this AI Roundup covered earlier this month has received generally positive reviews notwithstanding data privacy concerns.

• Sports gambling companies are using AI models to target users who are most likely to lose money, according to an investigation from The New York Times. Based on internal corporate memoranda and interviews with former data scientists at DraftKings, the investigation revealed that the sports betting operator developed algorithms specifically engineered to evaluate customer betting records and deploy promotional incentives toward those individuals likely to lose. Rather than relying on simple rule-based metrics, the company’s predictive models analyzed hundreds of account variables, including wager frequency, net loss ratios, daily balance swings, and churn probability, to generate an internal “elasticity” score measuring how much money a gambler was expected to lose for each promotional dollar received in bonus bets, profit boosts, or deposit matches. Bettors deemed “inelastic” were excluded from promotions, while marketing resources were directed towards high-elasticity users. While DraftKings maintains that promotional offers were directed towards users with “sustained, engaged” use and not based on losses, several former employees told The NYT that the algorithm was optimized to exploit vulnerable users. Furthermore, a parallel project to build models that would assign risk scores to users to detect early warning signs of problem gambling or addiction was shut down by DraftKings executives. The disclosures have intensified scrutiny from consumer protection advocates and state regulators, including the state of Massachusetts, who wish to crack down on predatory practices by sports gambling operators.