Welcome back to In the Loop, TIME new twice-weekly AI newsletter. We publish these episodes both as stories on Time.com and as emails. If you're reading this on a browser, why not subscribe to have the next article delivered straight to your inbox?
Who to know: Energy Minister Chris Wright
Last month, I interviewed Trump Energy Secretary Chris Wright for TIME's Person of the Year article: AI Architects. Wright, who comes from the private sector, is now betting on accelerating artificial intelligence. In our interview, he emphasized the role of AI in advancing critical scientific research and downplayed climate risks. These are excerpts of our conversation, lightly edited for clarity, with annotations.
How much of a priority is the AI race and building AI infrastructure for your mission?
This is the Trump administration's No. 1 science priority. We will be leaders in this effort, but this applies to the entire administration.
This is huge growth potential. Our national laboratories are applying AI to scientific discovery as quickly as possible. One important thing is that you can now take cancer or tumors and look at them and not just run tests and what they respond to, but actually understand their molecular structure. You can design molecules that can fight this ability to reproduce or grow. We hope that in the next few years, many cancers that are a death sentence today will become manageable conditions thanks to AI.
[Note: A group of oncologists wrote a paper about the limits of AI in cancer research, writing that while it is leading to “innovations in tumor detection, treatment planning, and patient management,” risks include “potential over-reliance on technology, which may undermine clinical expertise.”]
How important is it not to retire in old age? coal power plants your mission to power AI data centers?
Very important. We won't keep all of them open. But perhaps a large majority of all plants scheduled to close are closed for political reasons. Politicians and regulators agreed to shut down a coal-fired power plant 15 years before the end of its useful life. If we want to increase generating capacity as quickly as possible, we can stop digging a hole.
[Note: The U.S. Energy Information Administration (EIA) wrote in 2023 that many coal plants were being retired because they struggled to compete in the market with “highly efficient, modern natural gas-fired power plants and low-cost renewables.” But as AI has drastically increased demand, this calculus has changed.]
What will the energy mix be like when all these data centers are online?
Today, the largest source of electricity production in the United States is natural gas. So gas will definitely be the one that gets added the fastest. We are doing everything possible to develop traditional nuclear energy as quickly as possible. We will have many of these power plants built in the next 12 to 24 months, but it will be years before they become electrons on the grid.
Solar technology continues to evolve. I think we will see further development of solar energy even without subsidies. 33 years of wind power subsidies should be enough. These technologies should fly on their own. But they probably don't contribute much to new AI capabilities.
What would you say critics who claim that many wind and solar projects were launched, but funding cuts from the administration put them in jeopardy?
First of all, none of them have left now. We finally have legislative processes so the subsidies will end. If you start construction before July next year, you will receive these subsidies.
But then people say, “We need more electrons in the grid.” I tell people: There is no pool in the back where electrons are stored. We need more capacity during peak periods. That's all that matters. When it's the middle of the night and the wind is blowing really hard in Iowa, it doesn't increase our network capacity at all.
[Note: There are ongoing efforts to improve battery energy storage systems for renewables to address this problem. TIME has written about these efforts over the years.]
Sam Altman says he believes “over time, most of the world will be covered in data centers.” Do you support this vision?
I think they are extremely useful. The highest energy consumption is heat. So many people die from the cold. We changed the history of the planet by keeping people warm. Perhaps the second largest use of energy will be the creation of intelligence. So I think this is a great application.
Anything new that happens quickly causes public resistance, and for good reason. If the data center is part of your community and your electricity prices get up, you don't like it. I wouldn't like it either.
So we have a delicate balance. We have to build these data centers and supply them with energy and power, but stop the price increases. If hyperscalers need a gigawatt for their data center, they should give us a gigawatt of new capacity and pay rates that help stop prices from rising. This was not always done.
Climate activists complain that large corporations I'm walking back quietly their net zero goals. Does this concern you?
No. I think the public has a very unrealistic view of climate change, as if it is the biggest problem in the world today. If you look at the data on climate change, it doesn't even remotely resemble the world's biggest problem compared to hunger, public health, education and free trade.
[Note: The WEF’s Global Risks Report 2025 listed a variety of risks as more severe than climate-related ones over a two-year time period. But when considering a 10-year time period, all of their most severe risks were climate-related.]
Do you agree that this AI race has dramatically increased our need for energy production?
AI is contributing to the growth in electricity demand. The United States must gather its strength and begin to increase electricity production. But is this driving global energy demand? You know, not so much. But even if that were the case, would I still be for it? Absolutely. We are going to make people more efficient, more secure, more secure and more wealthy.
What You Need to Know: Fast AI Acceleration
AI capabilities in some areas are “doubling every eight months,” according to a major study released yesterday by the UK AI Safety Institute. A UK government body that conducts empirical research on advanced artificial intelligence models before their public release has recorded rapid progress in biological and cyber capabilities. The report also found that artificial intelligence companies' protections to prevent dangerous behavior by models are rapidly improving, with the time it takes experts to detect a jailbreak increasing from 10 minutes to more than seven hours.
Quick change – “We've seen really rapid improvement in capabilities in almost every area that the AI Security Institute measured,” Jade Leung, AISI's chief technology officer, told reporters on Wednesday. “In many areas, advanced systems now match or far exceed human experts.”
Loss of control – The AISI researchers also tested the models for “antecedent behavior,” which can indicate the model’s ability to escape the control of human operators, such as the ability to self-replicate, access the model’s own weights, and gain computational power. Based on a benchmark that included 20 of these challenges, model success rates increased from 5% in 2023 to more than 60% in summer 2025, according to the report. Leung said that all of these tests required specific prompting from the researchers and that no model exhibited spontaneous “relative” trends. Still, she said, “these metrics are important to track as we move toward more advanced systems over time.”
AI company – The report also includes new data on the extent of the relationship between human and artificial intelligence among UK adults. 33% of the more than 2,000 people surveyed said they used artificial intelligence models for “companionship, emotional support or social interaction” in the past year, including 4% who said they did so every day. Of these, the most common form of model was general purpose assistants such as ChatGPT, followed by voice assistants. Models created specifically for communication, e.g. Character.AIwas only 5%. — Billy Perrigo
AI in action
This week, Google quietly introduced its latest model, the Gemini 3 Flash, to Google Search and other platforms. Experts say it is smarter, faster and cheaper than previous models. It will be bad news for OpenAI if Google can reliably offer high-performance and low-cost models, since Gemini is already built into Chrome and Google Search.
What we read
“Podcast industry under threat as AI bots flood airlines with thousands of programs” Nilesh Christopher, Los Angeles Times
I've experienced this personally: I searched my podcast app to learn about a specific topic, selected the best option, and after just 10 minutes realized that I was listening to AI hosts based on the brutality of their banter and prose. Hundreds of thousands of AI-powered podcasts have flooded the apps, and “the invasion has just begun,” writes Christopher.
