AI News Roundup – SpaceX acquires AI coding agent Cursor amid IPO surge, AI systems help intercept drones in Russia-Ukraine war, new research finds AI search results can be easily manipulated, 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.

• Elon Musk’s SpaceX will acquire the company behind the Cursor AI coding agent as it seeks to compete for enterprise customers after its recent mega-IPO, according to Reuters. SpaceX will acquire Anysphere for $60 billion in an all-stock deal, gaining access to the latter company’s AI coding agent. SpaceX debuted on the Nasdaq index earlier this month, with its stock ending its first week of trading 37% above its IPO price of $135. The company’s market capitalization now stands at nearly $2.4 trillion, making it the sixth-largest company in the world. The company’s rapid rise made Musk the world’s first trillionaire. The Financial Times reported that SpaceX is planning to raise $20 billion in a bond sale in the coming weeks, partially to pay back a loan SpaceX took on when Musk merged his debt-laden AI company xAI into SpaceX. The Cursor deal is more evidence of the company’s drastic growth, and demonstrates the company’s desire to compete for enterprise customers, the market for which has been dominated by Anthropic and its Claude Code agent. OpenAI has also shifted its strategy towards enterprise customers, putting its Codex coding agent into higher focus (as this AI Roundup recently covered). SpaceX said that it will soon release an AI model on Cursor as well as Grok Build, its own coding agent developed jointly with Cursor in the past several months. SpaceX will reportedly pay a $4 billion to $10 billion termination fee if the Cursor deal falls through, depending on the circumstances, and it remains to be seen if the new deal will affect the company’s recent agreements with Anthropic and Google to lease cloud computing capacity to feed demand for both companies’ AI systems.

• The New York Times reports on the use of AI in the Russia-Ukraine war to intercept explosive drones. Since the start of the war in February 2022, the conflict has become a window into the capabilities of autonomous weapons systems that could define the future of warfare. Russia has used Shahed-type drones (which resemble small aircraft with onboard explosives) to bombard cities. Ukraine had previously relied on heavy machine guns and electronic means to combat Shahed drones, but a major Ukrainian drone manufacturer has developed a small interceptor drone that uses AI to help drone pilots detect and target Shahed drones before they reach their intended targets. Another company has reportedly developed an AI program that automates 95% of the drone interception process. Shahed-type drones have also been used by Iran in the ongoing war in the Persian Gulf region, and Ukraine has offered its interceptor drones (though without the AI technology) to other countries in the region. AI-assisted targeting systems are also being trained to identify and target humans. Ukrainian President Volodymyr Zelensky has said that these types of weapons heralded the future of warfare: “It’s only a matter of time — not much time — before drones are fighting drones, attacking critical infrastructure and targeting people all by themselves.”

• A recent experiment has shown that web search results presented by AI chatbots can be easily manipulated using Reddit posts, according to 404 Media. A recent paper from researchers at Cornell University found that AI systems that utilize web search, such as ChatGPT, often cite Reddit or Wikipedia for most queries. The researchers exploited this capability to make models output spam or scamming content using as little as a 13-word Reddit comment. Reddit and Wikipedia, both of which rely on volunteer moderators for policing content, have struggled to keep AI-generated content off of their platforms. Reddit, in particular, is prone to “shills,” accounts that do brand placement in the guise of normal comments. The new research has found that such efforts can greatly impact AI search results that many normal consumers have come to rely on, which hearkens back to traditional search engine optimization (SEO) tactics used by marketers since the dawn of internet search engines. One researcher told 404 Media that despite AI search changing how people find information online, AI systems often do not treat websites differently based on a source that may be more credible: “A random Reddit comment or an article from a government website. They are treated almost the same by the LLMs.” The very small amount of text needed to make these attacks work may make combatting them difficult, with the researcher saying that “there’s no easy fix.” In response to questions, a Reddit spokesman told 404 Media that “managing spam, bots, or other inauthentic content is not new to Reddit — we’ve been on the cutting edge of detecting and removing manipulated content and inauthentic accounts for 20 years. We have sophisticated systems that detect and prevent inauthentic behavior, coordinated manipulation, and astroturfing…”

• POLITICO reports on a new survey tracking global AI sentiment. The survey, conducted on 18,000 people across 15 countries by the U.K. research firm Public First, found that respondents in 11 of the 15 countries polled, including close U.S. allies like France, the U.K., Canada and Germany, now see China rather than the U.S. as the world’s leading AI superpower. The survey also revealed a growing backlash against AI in Western nations, particularly the U.S. and U.K., where net positive sentiment about the technology has dropped roughly 75% over the past two years. In the U.S., the share of people who believe AI will benefit society fell from 39% in 2024 to just 31% in 2026, while those expecting harm rose from 34% to 40%. The decline is sharpest among young Americans aged 18-24, whose opinion of AI’s societal impact swung from a 4-point positive margin to a 13-point negative margin this year. Americans’ top concerns regarding AI include misinformation, deepfakes, job displacement, and data center construction, on which the U.S. public is nearly evenly split, making it the most negative advanced economy surveyed on that issue. Meanwhile, countries like India, Singapore, and Vietnam remain broadly optimistic about AI. The findings land amid a lively debate in Washington over how much to regulate AI technology (as this AI Roundup has covered in recent weeks), with figures like former White House AI czar David Sacks warning that heavy oversight could result in ceding an advantage to China in the AI race. However, the survey results suggest the U.S. may already lack the public consensus needed to push the technology aggressively forward.

• Utah is planning to allow an AI system to renew prescriptions for patients and is sparking a backlash among physicians, according to The Wall Street Journal. The state is partnering with the startup Doctronic, which will allow Utah adults to log into a website to refill certain types of prescriptions, including antidepressants or cholesterol medication, though an AI system will perform the tasks usually done by human doctors. According to results released by the company and the state, the AI system approved 72% of renewal requests, escalating the others to human doctors. Doctors assigned to review the AI’s decisions agreed with the system’s determination 91% of the time. However, many doctors in the state have opposed the project, citing concerns about the risks of medication misuse and liability for patients who have issues. Most of the state’s medical licensing board signed a letter calling for the project to be suspended as the AI system hadn’t been properly vetted. Other researchers have raised doubts about Doctronic’s system due to a lack of studies on the AI’s decision-making or FDA approval for the product at all. In response to these concerns, a Utah official overseeing the project said that the company was required to obtain medical malpractice insurance, saying that despite the novelty of the process, “[Doctronic] need[s] to bear their ordinary responsibility of liability.” The project will not enter its next, more autonomous phase until the current trial has collected more data in the coming months.