AI News Roundup – OpenAI claims major mathematics breakthrough and sparks credit controversy, Claude used in warfare by US adversaries, Google DeepMind unveils AI tool for DNA research, 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 has claimed that its AI systems have solved the Navier-Stokes problem, though questions remain over proper credit for the breakthrough, according to the journal Science. The Navier-Stokes problem is one of six unsolved Millennium Prize Problems, a set of famous mathematical challenges created in 2000 by the Clay Mathematics Institute that each carry a $1 million prize for a verified solution. The Navier-Stokes problem asks whether the equations governing the flow of fluids like air and water will always produce physical, realistic solutions or if they can break down and mathematically “blow up” into a nonsensical result. Deploying an army of 10,000 autonomous AI agents over 88 hours at an estimated cost of millions of dollars in computing power, OpenAI generated a proof showing that under specific conditions, the equations produce a nonsensical, infinite result. However, this milestone ignited an intense dispute with academic mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge, who had spent months using OpenAI tools to make foundational breakthroughs on the problem. After OpenAI quickly deployed its massive compute resources to find a complete solution and publish before them, Buckmaster publicly raised concerns over whether OpenAI’s systems may have inadvertently drawn from their OpenAI user logs. Buckmaster alleged that OpenAI offered him authorship of the solution only if he excluded his Anthropic-affiliated collaborator, which Buckmaster refused. OpenAI initially claimed that “[w]hile unlikely, we cannot rule out that de-identified data derived from [Buckmaster and Alpöge’s] usage of our products helped improve our models,” though the company later said in a statement that “we have confirmed that Buckmaster’s Codex prompts…could not have influenced the system in any way, including through training.” The dispute has sparked broader anxiety in the academic community about confidentiality of inputs for some models and proper credit for AI-assisted work, as well as about high-tech corporations using their overwhelming resources to marginalize or displace human mathematicians.

• The Wall Street Journal reports that Anthropic’s Claude AI systems have been used at a large scale by U.S. adversaries in warfare. In a recent report, Anthropic described efforts by U.S. adversaries, including Iran, Yemeni Houthi rebels, and other groups based in China and Russia to develop weapons, surveil opponents, and distribute propaganda. In one instance, an Iranian group attempted to use Claude to identify vulnerabilities in U.S. Navy ships, while in another a Houthi group used the AI system to develop guidance for a ballistic missile system. These revelations come as the Houthi groups have recently gained control over the Bab al-Mandeb strait at the southern end of the Red Sea, a major channel for global trade. Anthropic said that the safeguards within Claude “blocked many of [the Houthis’] requests, but not all of them,” but that it banned the related accounts and shared intelligence information with U.S. authorities. The report also discusses the use of Claude by Chinese groups to surveil and compile intelligence on Uyghurs (an ethnic group in western China) as well as religious leaders throughout Asia. The company said that it hoped “that the findings in this report will help other developers recognize similar patterns on their own platforms, give governments and civil society a clearer view of how emerging threats take shape, and strengthen collective defenses.”

• Google DeepMind has announced AlphaGenome Atlas, an AI-powered search tool for research on human genetics, according to IEEE Spectrum. As this AI Roundup covered in July 2025, DeepMind at that time released AlphaGenome, an AI model designed to aid in genetics research by analyzing how changes to DNA sequences can affect biological functions. This past week, the company released AlphaGenome Atlas, a repository containing information on the effects of all nine billion possible single-letter changes to the human genome that were analyzed by the model, offering an easier way for scientists to make use of AlphaGenome’s capabilities without the need to run the model themselves. The predictions made by AlphaGenome have important limitations, as they only look at the adjacent one million base pairs of DNA around the change in question, while some changes could have much more wide-ranging implications to DNA elsewhere in the three billion base pairs in the human genome. However, the tool can still help scientists prioritize certain types of laboratory experiments and help accelerate research in new biological treatments, according to Žiga Avsec, the head of genomics at DeepMind. AlphaGenome access is freely available for non-commercial research, but the company is also considering commercial licensing opportunities.

• Bloomberg reports on the release of Meta’s customer-facing AI agent. The tool, dubbed “Muse,” was rolled out to adult users of Meta’s platforms in the United States this past week, and can assist users with shopping, scheduling appointments, and communicating with other people through integrations with Meta’s social media apps including Facebook, Instagram, and WhatsApp, as well as third-party tools such as Ticketmaster, OpenTable, and Google’s web services suite. The release of Muse is in line with Meta CEO Mark Zuckerberg’s AI vision, which he outlined in a manifesto last month (as this AI Roundup covered at the time), that envisioned a future where everyone had access to “an exceptionally capable personal agent that understands you, your goals, and everything you care about.” Meta’s Chief AI Officer Alexandr Wang told Bloomberg that Muse is “an extremely secure system,” is independent from Meta’s advertising systems, and is designed to ask for user permission before making purchases or other decisions on sensitive matters. Muse will be offered in both free and paid versions, with the latter offering access to more computing power.

• Public schools in New York City are banning AI usage for K-8 students, according to WNYC. The one-year moratorium was announced earlier this month by New York Mayor Zohran Mamdani and reverses guidance from earlier this year that allowed some limited uses of AI. That guidance received pushback from students’ parents. New York City Schools Chancellor Kamar Samuels told WNYC that the decision was made in part to encourage young students to do more work themselves rather than outsourcing that work to AI chatbots, saying that students “learn best when they are wrestling with hard problems, when they’re getting it wrong, when they’re trying again and finally getting it right.” In a related announcement, NYC schools will also introduce limits on screen time for young students using individual laptops or tablets in the third through eighth grades, while students in kindergarten through second grade won’t be able to use such devices at all in the classroom absent a disability requiring such technology. Teachers will also be prohibited from using AI tools for grading, discipline, and decisions on student advancement, though the local teachers’ union has questioned that policy. Immediate reactions to the new technology and AI policies from parents have been mostly positive, though some advocates want the AI ban extended through high school and beyond the one-year period.