AI News Roundup – Anthropic and the Pentagon at odds over military AI use, DeepMind developing model to study human genome, MUFG deploys “AI employees,” 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.

• • Reuters reports that AI developer Anthropic and the U.S. Department of Defense (DoD) are sparring over the proper deployment of the former’s AI technology. Anthropic and the DoD signed a $200 million agreement in July 2025, but the parties have been at odds over the amount of safety guidelines and human oversight that should be implemented into the technology. Anthropic reportedly raised concerns that its AI tools could be used to surveil American citizens or be used for weapons targeting without enough human oversight. Anthropic claims to have guided its AI development around AI safety concerns, and its models, including those in government applications, are trained accordingly. The Pentagon has pushed back on these concerns, arguing that they should be able to use AI in any way that does not violate U.S. law, regardless of a vendor’s usage policy. There appears to be no immediate off-ramp to the current impasse between the company and the DoD, but one source told Reuters that the Pentagon will likely need Anthropic’s cooperation to move forward and retool its models if necessary. 
  • Google DeepMind is developing an AI model to aid the study of the human genome, according to The New York Times. The new AI program, AlphaGenome, was trained on molecular data, and can make predictions about genes and mutations that could lead to cancer and other diseases, and thus help identify possible treatments for those conditions. In 2024, DeepMind scientists shared the Nobel Prize in Chemistry for their work on the AI program AlphaFold, which predicted the shapes of proteins and can aid in the production of new ones and development treatments for disease. AlphaGenome is thus seeking to do something similar with DNA. Some experts expressed caution about the new model, saying that it “is not going to win the Nobel Prize,” but that it was a useful tool for continuing research into DNA and treatment of genetically-caused conditions. 
  • Nikkei Asia reports that Mitsubishi UFJ Financial Group (MUFG) is deploying “AI employees” to work alongside human ones. MUFG, one of the world’s largest banking conglomerates, announced plans to have AI assistants work alongside employees as soon as this month. These assistants will be able to perform tasks such as writing speeches and onboarding new hires. The AI assistants were developed in-house, though some are based on models developed by OpenAI, and will go through a similar onboarding process to new human hires, going through an initial phase of training on tasks before being placed in a department and improved through contact with employees at the company beginning in April. 
  • New research has uncovered one possible method to reduce AI energy use: thermodynamic computing, according to an analysis by IEEE Spectrum. Thermodynamic computing uses physical circuits that change state in response to random noise, such as thermal fluctuations, to perform low-energy computations. Earlier this month, researchers at the Lawrence Berkeley National Laboratory reported that they had created a neural network using thermodynamic computing methods. The process involves providing a set of images to a thermodynamic computer, which would then randomly degrade. This process would then be reversed using a probabilistic method. Simulations of this process have demonstrated that such thermodynamic computers can generate images of handwritten numerical digits. If scaled up, this technology may allow for image generation — a key application of AI — that uses much less energy than current methods using traditional computers. 
  • Bloomberg reports that the U.K. is pledging to offer free AI training to all workers to boost economic output. Technology Secretary Liz Kendall announced this past week that every adult in the U.K. will be eligible to take free online courses that will teach them how to use AI tools for “drafting text, creating content and completing administrative tasks,” with a goal of freeing up time for other work. The U.K. government has prioritized AI as it seeks to boost the country’s economy, though many still remained concerned about AI’s ability to displace workers entirely rather than assist their productivity. Recent research has shown that the U.K. is losing more jobs due to AI than creating them, but Kendall has said that the Department for Science, Innovation and Technology is addressing those concerns, saying that, “Change is inevitable, but the consequences of change are not. We will protect people from the risks of AI while ensuring everyone can share in its benefits.”