Top Story
Trends from Stanford’s State of AI 2026
Each year, Stanford’s Institute for Human-Centered AI releases a report of the state of artificial intelligence based on the events of the past year. This year, Stanford HAI identified 15 overarching trends that are shaping the present and future of artificial technology globally. Here are some of the most interesting:
AI capabilities are accelerating, and reaching more people than ever. Multiple AI models now meet or exceed human baselines in areas such as answering PhD-level science questions, math competitions, and computer coding. Adoption of AI continues to increase - 88% of large organizations use it in some form, as do 4 out of 5 university students.
AI capabilities still exhibit a “jagged frontier”. AI models can win math Olympiads, but still can’t reliably tell time from an old-fashioned clock with hands.
The US still leads in AI investing, but its ability to attract global talent has cratered in the past year. US private investment in AI reached nearly $300 billion in 2025, more than all other countries combined. Yet the number of AI researchers and developers moving to the US declined 80% in the past year, which should be no surprise to anyone.
Only 31% of Americans trust their government to regulate AI effectively, the lowest proportion in an Ipsos survey of 30 countries. This is compared to a global average of 54% trust. In Singapore, 81% of respondents expressed trust in their government to effectively regulate AI.
US scores dead last among 30 countries surveyed for citizen trust in government to regulate AI.
Clash of the Titans
Feds beg Anthropic to come back, NSA is using Claude despite ban
Two months after the well-coifed Secretary of Defense banned Anthropic from the Pentagon and labeled it a “supply chain risk”, numerous federal agencies are still using, and even increasing their use of, Anthropic’s Claude AI model.
The reasons are simple. First, Claude is an extremely good model. Second, it is already embedded in workflows in many agencies, and scrapping all that to switch to OpenAI’s ChatGPT or Elon Musk’s Grok is a lot of work for no gain in function. Third, Anthropic just scared the bejeesus out of the whole world two weeks ago when it announced its new Mythos model, which is apparently so insanely good at coding that it is a superhacker in a box. Mythos is apparently able to discover and exploit vulnerabilities in software that humans have vetted and deemed safe for decades. Mythos is so scary good at cyberhacking that Anthropic only released it to 40 organizations, so that they could use it to harden their codebases before general release of the model.
Mythos is so scary good that the NSA spy agency is openly using it despite the ban. US Director of National Intelligence Tulsi Gabbard is in effect saying to Hegseth “fight me” (or more likely, “bite me”) if he wants to take away the world’s best tool for cyberhacking (an NSA specialty against the bad guys) and conversely, cybersecurity (to help the good guys.)
So, like the ex that dumped you calling tearfully to ask “Can we talk?”, Trump’s Chief of Staff Susie Wiles invited Anthropic to the Oval Office last Friday for a confab. Afterward, officials called the meeting “productive and constructive.” We await, as they say, further developments with interest.
Anthropic CEO Dario Amodei meets with Trump Chief of Staff Susie Wiles while Hegseth fumes.
Dem Congressional candidate pushes for “AI Dividend”
Alex Bores, 35-year-old Democratic Assemblyman from New York, is running for Congress on a platform focused on AI safety and protection of workers.
The boldest proposal in Bores’ platform is the “AI Dividend”, a direct payment to Americans, funded both by taxing AI companies on usage, and through having the federal government establish a sovereign wealth fund that holds equity shares in AI companies. This fund, similar to Norway’s $2 trillion fund established to hold the profits from its North Sea oil reserves, would be used to fund government operations involved in overseeing AI, as well as direct payments to US residents. The state of Alaska has a similar program of paying residents an annual dividend funded by the state’s oil revenues.
AI billionaire political action committees are pouring money into opposing Bores, who is co-sponsor of New York’s landmark RAISE act, an AI safety bill. Bores points out the irony that his proposals merely take AI CEOs at their word that AI can pose huge safety hazards, and may make all jobs obsolete. His proposals for a tax on AI, and a sovereign wealth fund that holds equity in the AI companies were prominently featured in OpenAI CEO Sam Altman’s recent 13-page think piece entitled “Industrial Policy for the Intelligence Age: Ideas to Keep People First.”
Bores is no Luddite. In fact he came of age by working in the belly of the AI beast. He earned a Bachelor of Science degree at Cornell, then a Master’s in computer science at Georgia Tech. Afterward he worked as a software engineer at Palantir, the defense technology company whose Maven AI system is currently making targeting decisions in the war in Iran.
He has seen AI and the people who develop it, up close and personal, and he has concluded that we should take AI CEOs at their word on the the promise and dangers of AI, and and that we should hold them accountable to turn their gauzy think pieces on how to achieve a future of abundance for all into hard reality.
Bores is a new father as well as a candidate for Congress.
Fun News
Palantir releases “completely normal” manifesto
Alex Karp, CEO of defense AI company Palantir, and to all appearances a deeply weird human being, last year penned a book (or maybe had an AI chatbot write it) called the “Technological Republic.” Reception was mixed, with Silicon Valley Tech Bros and Wall Street billionaires praising it, and most other people giving it a “meh.”
Now Palantir, the same company whose Maven AI system is running the war in Iran, the same company that caused the tussle between Secretary of Defense Pete Hegseth and Anthropic CEO Dario Amodei over rogue killer robots, THAT Palantir, has released a 22-point manifesto that is a distillation of the main ideas in the book.
Stripped down to their essence in bullet-list form, the ideas in that book are not a little disturbing, coming from a major defense contractor and major AI company.
It is a chest-thumping, anti-woke, Western supremacist, militaristic, and profoundly anti-democracy screed. Elliot Higgins, CEO of renowned investigative journalism organization, Bellingcat, dryly noted that it was “Extremely normal and fine for a company to put this in a public statement.” He went on to say that the manifesto was an “attack on the key pillars of democracy.” Other critics called it “a supervillain’s manifesto.”
Just another day at the office for Palantir, I guess.
Palantir CEO Alex Karp.
Data center development costs dwarf other megaprojects
When describing the current massive investment in data centers for AI, it’s hard to grasp just how big it really is. Below is a chart depicting the amount of investment in other US megaprojects over the lifetime of the project.
What becomes instantly clear is that the AI buildout is consuming vastly more capital in significantly less time than any other comparable project. Data center capital expenditures have totaled $930 billion in just 6 years, with hundreds of billions more planned for 2026 alone. As a comparison, in inflation-adjusted dollars:
The Manhattan Project that created the atomic bomb in World War II consumed $35 billion over 5 years
The Marshall Plan that revived Europe after that war took $170 billion over 4 years.
The US Space Program that put a man on the moon consumed $257 billion over 14 years.
Building the nation’s railroad system took $550 billion over 71 years, and
Building the Interstate Highway System required $620 billion over 37 years
So not only the scale, but the velocity of the spending is unmatched in modern times.
And for a fun comparison, how do revenues of AI companies compare with the revenues from the current blockbuster GLP-1 weight loss drugs?
(Answer: Until this year, the revenues were not even close, with GLP-1s generating$30-$40 billion a year each, and AI revenues a fraction of that. Next year, both OpenAI and Anthropic are expected to exceed GLP-1 revenue for the first time.)
Robots
Robot beats human record in Beijing half marathon
Last year, at the so-called Robot Olympics in Beijing, humanoid robots struggled to stay upright, fell over a lot, and shuffled at a snail’s pace for 100 meters.
What a difference a year makes.
On Sunday, a humanoid robot beat not only all 12,000 humans entered in the Beijing Half-Marathon, it beat the existing human world record time by nearly 7 minutes.
Pictured below, the 5 feet-5 inch robot, named “Lightning”, completed the half marathon in 50 minutes, 26 seconds. The human world record, set only 6 weeks ago by Ugandan runner Jacob Kiplimo, is 57 minutes, 20 seconds.
“Lightning” is a research prototype built by smartphone manufacturer Honor, and includes innovative liquid cooling based on smartphone manufacturing designs.
Honor’s “Lightning” humanoid robot lives up to its name, besting the human world record time.
Robotic legs help elderly Hong Kong fire survivors climb stairs
Months after a deadly fire gutted an apartment block in Hong Kong, survivors are being allowed to climb the stairs of their former high-rise home top retrieve their belongings. For many of the 1,500 elderly residents, a climb of up to 31 stories is just not possible without assistance.
An NGO named AidVengers Federation has stepped in and made robotic exoskeletons manufactured by Shanghai firm Hypershell available to assist elders who pass a basic stability and proficiency test. So far, about 70% of applicants have passed. The battery-powered devices are strapped onto the legs. When the user makes a motion to lift their leg, an electric motor helps lift and then straighten the leg to assist with stair-climbing. Such devices are becoming common in China for sightseers who want to climb long stairs or mountain trails.
Using the robotic exoskeleton, elderly residents are able to make multiple trips to their apartments on high floors, to bring back valuables and sentimental objects such as family photos.
Robotic exoskeleton helps users climb stairs, or even mountains.
AI in Medicine
AI helps cardiologists image the inside of the heart during procedures
Cardiac ablation for atrial fibrillation is a minimally invasive procedure that destroys small areas of abnormal cells in the heart to interrupt aberrant electrical activity that prevents normal heart rhythm. It requires utmost precision so that only the abnormal cells are destroyed, but precision is difficult when the physician is at the long end of a 5-foot catheter and can’t easily see exactly where the catheter tip is.
Now Johnson and Johnson has introduced their Cartosound Sonata system. It uses AI to automatically convert echocardiogram images from inside the heart into 3-D maps of the organ, with all landmarks automatically labeled. This makes ablation less like fumbling in the dark with a match and more like driving a car in the daylight.
AI maps the heart in real time - from the inside.
OpenAI and Anthropic cozy up to Big Pharma
AI companies are targeting all the major industries, looking to land customers able to fuel their massive cash burns. Recently, both OpenAI and Anthropic have made moves to consolidate relationships within a very lucrative sector of the healthcare industry, namely Big Pharma.
OpenAI has recently announced a strategic partnership with NovoNordisk, maker of the Ozempic/Wegovy GLP-1 weight loss drugs, to integrate its AI models into drug discovery, as well as help optimize manufacturing, distribution, supply chain management, and corporate operations. Both partners believe that AI can significantly speed up the time from initial research to patent application.
Meanwhile, Anthropic has gone even further with the chumminess, and added Vas Narasimhan, the CEO of Novartis, to its Board. Novartis is perhaps best known as the maker of Entresto, the heart failure drug. Narasimhan, in a prepared statement, said that he was most interested in seeing AI developed ethically and responsibly. He joins a star-studded Anthropic Board, which contains both Reed Hastings, co-founder of NetFlix and Chris Liddell, former CFO of Microsoft and of General Motors.
Novartis CEO Vas Narasimhan at Davos.
That's a wrap! More news next week.