AI News Roundup – OpenAI discontinues Sora app and shifts priorities, Arm to develop AI CPUs, study warns against sycophantic AI behavior, 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. 

• OpenAI is discontinuing its Sora video-generation service as it shifts its corporate strategy, according to The New York Times. Sora was first unveiled in 2024, and was quickly noted for its ability to create videos featuring many copyrighted characters. Several rightsholders, including the Walt Disney Company, signed licensing agreements with OpenAI to allow for their IP to be used in Sora outputs, but Disney has since withdrawn from such agreements and its related OpenAI investment commitments. While the Sora consumer app was initially popular when it was released in 2025, it never reached the popularity of ChatGPT, and a lack of monetization coupled with the heavy computing cost of AI video generation led to the recent move to shut down the service. OpenAI has faced pressure as it plans an initial public offering (IPO) later this year to generate more revenue, and cancellation of Sora is not the only sign of the company’s shifting priorities; the Financial Times reported this week that the company has placed its plans for a chatbot that produces erotic material on hold indefinitely, following pushback from within and outside of the company. The company is expected to focus on more business-focused products, such as its Codex programming system. Regardless, OpenAI is still expected to move forward with its IPO in the second half of 2026.

• WIRED reports on Arm Holdings’ announcement that it is developing its own AI semiconductors. At an event in San Francisco, Arm CEO Rene Haas announced that the company would produce its own AI-focused CPUs, a departure from the company’s prior business model of licensing chip designs to manufacturers. The new chip is dubbed the Arm AGI CPU, and is intended to be coupled with other processors in data centers for AI tasks. Taiwan Semiconductor Manufacturing Corporation (TSMC), the world’s leading computer chip producer, will manufacture the new chips for Arm. Arm has claimed that its new CPUs are much more energy-efficient than competing offerings from Intel and AMD, and that it already has several major customers lined up, including Meta, OpenAI, and the Internet infrastructure company Cloudflare. Further details on the chip are expected later this year.

• A new study has found dangers in the flattering and sycophantic way AI chatbots behave towards their users, according to the Associated Press. The study, published in the journal Science, was conducted by researchers at Stanford University and tested 11 major AI systems. The research found that all models demonstrated overly agreeable behavior, which has sparked concerns over the effects of such behavior on users, who are more likely to trust AI systems when they confirm their priors. The study also pointed to this behavior as a possible source of high-profile cases of delusional behavior in some users, where the AI system convinced them of impossible things. The researchers were especially concerned about this behavior’s effect on children and young adults, as reductions in social friction and conflict could impair their cognitive and social development. Addressing this issue in chatbots could require retraining and restructuring them from the ground up; as one researcher said, “we want AI that expands people’s judgment and perspectives rather than narrows it.”

• Block Club Chicago reports that a private school with AI teachers will be opening in the city this coming fall. Alpha School, which has locations around the U.S., is opening a school in Chicago’s Lakeshore East neighborhood in fall of 2026. While the locations are human-staffed, instruction is performed through an AI-driven software platform in only two hours per day, with the rest of the school day dedicated to projects or group work. The Alpha School has a tuition rate of $55,000 per year, placing it among the most expensive schools in the city and is planning to serve 100 students in kindergarten through the eighth grade in its first year. The use of AI in education has been controversial, especially as research has warned of the dangers of AI use on general cognition, and the local Chicago Public Schools district has a much more limited use of AI. One Chicago school board member told Block Club that using AI to teach a curriculum made her “very nervous.” Alpha Schools’ nationwide enrollment has grown from 200 to over 1,000 over the past few years, and the network has been praised by Trump administration officials as “exemplary” and a herald of the future of education.

• Apple will be opening up its Siri assistant to use non-ChatGPT AI models in its next major iOS update, according to Bloomberg. Since Apple’s AI offerings launched in 2024, users have been able to send Siri queries to ChatGPT, but the iPhone maker is reportedly developing features for iOS 27 that will allow AI chatbots beyond those from OpenAI to integrate with Siri, including Google’s Gemini and Anthropic’s Claude. These plans are reportedly separate from Apple’s arrangement with Google for Gemini models to power Siri, as this AI Roundup covered earlier this year. The new features, dubbed Extensions, will connect Siri users directly to the iOS apps for the third-party AI services, and certain services can be enabled and disabled at will. Apple is expected to announce iOS 27 at its Worldwide Developers Conference this June.