Artificial Intelligence

[phkrause]

How to AI

Illustration: Brendan Lynch/Axios

The No. 1 pushback we get from AI skeptics or newbies is: "It's overhyped. I asked it something and it spit out an unimpressive answer!" Truth bomb: It's not the AI. It's you, Jim VandeHei and Dan Cox write. Why it matters: Ask top large language models like ChatGPT something simple or generic, and you will get a simple or generic answer. Ask the right questions the right way, and you will often get magic. Jim asked Dan Cox, our CTO and AI leader, to help him craft six ways for ordinary users to get more extraordinary answers. It starts with prompts — the very questions you pose to ChatGPT, Claude, Gemini, Grok or Perplexity. Just in time for performance-review season, here are six tips (for both reviewers & reviewees): 1. 🔁 It's a conversation, not a search engine. The biggest mistake newbies make is treating AI like Google — one question, one answer, done. The magic happens in the back-and-forth. Ask a question. Read the answer. Then push: "Make it shorter ... Give me three alternatives ... That's too formal ... What am I missing?" The best outputs come from the fifth or sixth exchange. Example: You ask for help requesting a raise. The first draft is generic. You say: "Too corporate. I've been here six years and my boss is informal — make it sound like a real conversation." Now it's useful. 2. 👤 Nail the Who. Start by explaining who you are — your role, experience, anything relevant — and who you want the AI to think like when answering. Be specific. Example: "I'm a senior account manager at a midsized software company. I've been here six years, consistently hit my numbers, and just took on two new direct reports. I want you to think like a brutally honest executive coach who has helped hundreds of people negotiate compensation." 3. 🧩 Context matters most. Give detailed, real-world framing upfront. The AI doesn't know your situation, your audience, or what you're trying to avoid unless you tell it. Specificity is everything. Example (building on the Who😞 "My annual review is in two weeks. I haven't had a raise in 18 months despite a promotion in title. I know the company had a rough Q3, but my division exceeded targets. My boss is supportive, but not the final decision-maker — he has to pitch it to the VP. What's the smartest approach?" 4. 🚫 Just say no. Tell it what not to do. This sharpens output dramatically. Example (adding constraints): "Don't give me generic advice like 'know your worth.' Don't suggest ultimatums — I'm not bluffing. And don't make it sound like a script I'd read verbatim." 5. 🪜 Say: "Think step by step." When you're dealing with anything complex — a negotiation, a decision with trade-offs, a strategy with multiple variables — ask the AI to reason through it explicitly. This simple phrase dramatically improves output quality. Example: "Think step by step about how my boss will react and what objections he might raise when pitching this to the VP." 6. 👀 Just dump the image in. The models are extraordinary at instantly understanding screenshots, documents or files. Stop wasting time explaining what you're looking at — the AI can just see it. Example: Screenshot your company's salary bands from the internal HR portal. Paste it into ChatGPT alongside your title and tenure. Ask: "Based on this, where should I be? What's a reasonable ask?" ⚠️ Trust but verify. AI can hallucinate confidently — inventing facts, statistics, even citations that don't exist. The more specific the claim, the more you should double-check. Use it to think, draft and strategize. But if it spits out a number or a name, verify it before you repeat it. The bottom line: AI is a power tool. It rewards users who treat it like a sharp colleague rather than a magic box. Be specific. Be demanding. Keep pushing. 📱 How'd you do? How'd we do? What's your power prompt? Let us know: finishline@axios.com. Go deeper: Jim's video, "Blunt AI advice."

[phkrause]

Building AI brains

Illustration: Allie Carl/Axios

AI startups are raising billions to develop "brains" for robots that could work everywhere from oil rigs to construction sites, Axios Pro Rata author Dan Primack writes. Why it matters: Blue-collar workers may have as much to fear from AI disruption as white-collar workers. These software "brains" would understand physics and real-world conditions — helping the robots adapt to changing environments. Some of these AI-powered robots may be humanoids, others may not — form matters less than functionality. Plumbing, electrical, welding, roofing, fixing cars, making meals — there really isn't much of a limit. Think about it a bit like C-3PO and R2-D2, but without the snarky personalities. 🔬 Zoom in: There isn't yet agreement on the smartest way to apply AI to robotics. Big Tech giants and startups are gathering gobs of real-world data to train their AI models. Others employ "world models," which are trained on simulated physical world data. They're cheaper — relying on an understanding of things like gravity — and have been championed by Yann LeCun, the former chief AI scientist at Meta who recently formed a new company called AMI Labs. Follow the money: Toronto-based Waabi last week raised up to $1 billion in what could be the largest funding ever for a Canadian startup, with an initial focus on robotaxis and self-driving trucks. Pittsburgh-based Skild AI just raised $1.4 billion at a $14 billion valuation. Its motto: "Any robot. Any task. One brain." FieldAI last month raised nearly $400 million to focus on "dirty, dull, or dangerous" industries like energy and logistics. Its software could help robots build data centers — AI enabling AI, leaving humans on the sidelines. 🥊 Reality check: It's impossible to know how many blue-collar jobs could be rendered irrelevant, or over what time frame, as AI expands from the virtual to the physical. Even if an AI-powered robot can outperform a human, the added hardware and switching costs may outweigh the added efficiency. At least for now.

[phkrause]

🤠 AI's first rodeo

Image: TWG AI/Teton Ridge/Palantir AI is moving into rodeo arenas and bringing analytics to one of America's most tradition-bound sports, Axios' Russell Contreras writes. Why it matters: Rodeo has long defined itself as the last major American sport untouched by analytics. If AI takes hold in training and broadcasting, it won't just change how riders compete but redefine the cowboy's identity. Cowboys are one of America's most enduring symbols of independence and tradition. Their embrace of AI could serve as a bridge between high-tech innovation and communities that often see it as threatening. 🐎 Zoom in: Palantir, TWG AI, and Nvidia announced last month they are teaming up with Texas-based Teton Ridge to bring real-time AI and computer vision into rodeo arenas. Keep reading.

[phkrause]

🏎️ The other AI race

Illustration: Brendan Lynch/Axios Anthropic has signed a multiyear deal with racing's Atlassian Williams F1 Team, Axios' Eleanor Hawkins reports. 🧠 Anthropic's Claude will be the team's "Official Thinking Partner." The team's cars will get new Anthropic branding. 🌎 Lots of AI companies are trying to tap into F1's huge global fandom. Oracle has teamed up with Red Bull Racing since 2021. Google is the official partner of McLaren's F1 team. IBM locked in a partnership with Scuderia Ferrari, and Perplexity signed driver Lewis Hamilton as a brand ambassador. Microsoft recently announced a multiyear deal with Mercedes-AMG Petronas F1 Team. Go deeper.

[phkrause]

AI's software disruption is here

Illustration: Brendan Lynch/Axios

Software stocks are getting dumped as investors price in a world where AI could replace software services, with the selloff dragging down the broader market yesterday. Why it matters: The wipeout shows how the market can respond when facing fresh evidence that AI could replace an entire industry, Axios Markets' Madison Mills reports. 📉 Software stocks slid after Anthropic rolled out new AI automation capabilities for several different sectors of enterprises. Selling started in legal software/data-adjacent names, including Experian, the London Stock Exchange Group, Thomson Reuters and LegalZoom, then broadened across the sector. The iShares Expanded Tech-Software Sector ETF (IGV) is down more than 14% over the past six sessions, following a 15% drop in January (its worst month since 2008). Software sentiment is the "worst ever," investment firm Jefferies says in a blunt note. The category is "radioactive," Anurag Rana of Bloomberg Intelligence tells Axios. 🦾 The backstory: Anthropic sees itself as a complement rather than as a competitor to software providers. Anthropic can securely connect with other tools and applications, becoming your "home page" of sorts while software services run through the back end. Claude can "render interfaces directly within it," potentially driving "even more engagement and interactivity … with all these other business systems," Scott White, head of product for enterprise at Anthropic, tells Axios. Reality check: It's not the first time Wall Street has turned sour on software. Mobile was once expected to threaten Microsoft's software business, since everyone was going to be spending time on phones, not desktops. Microsoft's stock is up 789% over the last decade. 🤖 Nvidia CEO Jensen Huang, speaking in San Francisco at an AI conference hosted by Cisco, called the idea of AI replacing software "the most illogical thing in the world." "If you were a robot," Huang said, "would you use tools or reinvent tools? The answer, obviously, is to use tools ... That's why the latest breakthroughs in AI are about tool use."

[phkrause]

• 💻 OpenAI today released a new enterprise platform, Frontier, designed to let large companies build, deploy and manage fleets of AI agents that plug into their existing systems. Go deeper.

[phkrause]

Real or AI? 7 Clues to Spot Fake Images Right Away

AI-generated images are getting scarily realistic, but there are still clear signs to help you spot the fakes.

https://www.pcmag.com/explainers/real-or-ai-7-clues-to-spot-fake-images-right-away?

[phkrause]

⚖️ Legal world enters AI era

Illustration: Brendan Lynch/Axios

AI promises to make lawyers more productive, but there's a problem: Their clients are using it, too, Axios' Emily Peck writes. Why it matters: AI is creating new headaches for attorneys. They're worried about the fate of the billable hour — a reliable profit center for eons — and are perturbed by clients getting bad legal advice from chatbots. Dave Jochnowitz, a partner at the law firm Outten & Golden, says it's "like the WebMD effect on steroids," referring to how medical websites can give people a misguided impression. "ChatGPT is telling them: 'You got a killer case,'" Jochnowitz said. But the models don't always understand the full context, applicable laws or case history. 👓 Between the lines: The potential impact of AI on the industry was evident when shares of legal software companies, like LegalZoom and Thomson Reuters, fell sharply after Anthropic released a legal plug-in.

[phkrause]

🔌 AI fuels massive power demand surge

Data: International Energy Agency. ("Other" includes cooling, heat pumps and other building needs.) Chart: Kavya Beheraj/Axios Data centers are slated to account for 50% or so of U.S. power-demand growth for the rest of the decade, Axios' Ben Geman reports from a new International Energy Agency analysis. The AI-driven rise of huge data centers is a big reason IEA sees overall U.S. demand rising 2% annually on average from 2026–30. That's twice the pace from 2016–25. Keep reading ...

[phkrause]

🥊 AI infighting hits boiling point

Illustration: Annelise Capossela/Axios

AI CEOs are openly trash-talking each other, sniping over advertising and their philosophical approaches to the future, Axios AI+ co-author Madison Mills writes. Why it matters: The squabbling is intensifying as the cost of staying competitive in AI soars, along with pressure to deliver real returns. 🖼️ The big picture: AI CEOs can be divided into two groups — the researchers and the entrepreneurs. The researchers tend to view AI as a fragile, long-term project that demands collaboration, caution and governance. Google DeepMind's Demis Hassabis and Anthropic's Dario Amodei fall into this camp. The entrepreneurs — including OpenAI CEO Sam Altman and xAI's Elon Musk — want to move fast and break things. The fighting ramped up when Anthropic pledged to keep its large language model Claude ad-free and ran a Super Bowl commercial poking at OpenAI, which is testing ads in ChatGPT. Altman fired back with a 420-word post calling the ad "dishonest." Altman also has beef with Musk. Musk is pursuing two lawsuits against Altman. Keep reading.

[phkrause]

👀 OpenAI's ad test begins

Illustration: Sarah Grillo/Axios

ChatGPT started testing ads for some U.S. users yesterday on its free and cheapest subscription tiers, Axios' Madison Mills writes. Why it matters: It could be the beginning of the end of ad-free ChatGPT. The company says ads won't influence ChatGPT's answers, which will remain focused on what is "most helpful." If a user asks about recipe ideas, the answer may be followed by a grocery delivery service ad, for example. Keep reading.

[phkrause]

Loud AI alarm

Illustration: Allie Carl/Axios

Top AI experts at OpenAI, Anthropic and other companies warn of rising dangers of their technology, with some quitting in protest or going public with grave concerns, Axios' Madison Mills writes. Why it matters: Leading AI models, including Anthropic's Claude and OpenAI's ChatGPT, are getting a lot better, a lot faster, and even building new products themselves. That excites AI optimists — and scares the hell out of several people tasked with policing their safety for society. On Monday, an Anthropic researcher announced his departure, in part to write poetry about "the place we find ourselves." A former OpenAI researcher, Zoë Hitzig, also left this week, citing "deep reservations about OpenAI's strategy." Another OpenAI employee, Hieu Pham, wrote on X: "I finally feel the existential threat that AI is posing." Jason Calacanis, tech investor and co-host of the All-In podcast, wrote on X: "I've never seen so many technologists state their concerns so strongly, frequently and with such concern as I have with AI." The biggest talk among the AI crowd yesterday was entrepreneur Matt Shumer's post, "Something Big Is Happening," comparing this moment to the eve of COVID. It went mega-viral, garnering nearly 70 million views in 36 hours, as he laid out the risks of AI fundamentally reshaping our jobs and lives: "This might be the most important year of your career. Work accordingly." 🚨 Between the lines: While the business and tech worlds are obsessed with this topic, it hardly registers in the White House and Congress. 🥊 Reality check: Most people at these companies remain bullish that they'll be able to steer AI smartly, without societal damage or big job loss. But the companies themselves admit the risk. Anthropic published a "Sabotage Risk Report" showing that, while low risk, AI can be used in heinous crimes, including the creation of chemical weapons. The report examined risks of AI acting without human intervention, purely on its own, including "self-sustaining activities that allow it to pay for or steal access to additional compute." OpenAI dismantled its mission alignment team, which was created to ensure AGI (artificial general intelligence) benefits all of humanity, Platformer's Casey Newton reported. 💥 Threat level: The latest warnings follow AI breakthroughs showing these models can build complex products on their own — then improve them without human intervention. OpenAI's last model helped train itself. Anthropic's viral Cowork tool built itself. These revelations — in addition to signs that AI threatens big categories of the economy, including software or legal services — are prompting lots of real-time soul-searching. The bottom line: The AI disruption is here. Its impact is happening faster and more broadly than most people and institutions are ready for.

[phkrause]

🦾 Anthropic subscription surge

Data: Ramp AI Index. Chart: Axios Visuals The share of U.S. companies paying for Anthropic's AI tools and services jumped in January, Axios' Emily Peck writes. Why it matters: It was Anthropic's breakthrough month, writes Ara Kharazian, an economist at Ramp, which has been tracking business spending on AI. Anthropic's new software coding product — Claude Code — went viral earlier this year and helped drive adoption. 🧮 By the numbers: The share of companies paying for Anthropic increased to 20% from 17%, according to Ramp, a company that offers corporate credit cards and expense-management tools to roughly 50,000 companies nationwide. OpenAI dropped slightly from 37% to 36%. One-fifth of businesses that use Ramp now pay for Anthropic, up from one in 25 last year. Anthropic isn't gaining users at OpenAI's expense. About 79% of OpenAI users also pay for Anthropic.

[phkrause]

💵 Anthropic's $30 billion bonanza

Illustration: Sarah Grillo/Axios

Anthropic raised $30 billion in one of the largest private funding rounds in tech history, Axios's Madison Mills and Dan Primack write. Why it matters: Investors are eager to pour billions into an AI race that is heating up faster than even the optimists could have imagined. 📖 By the numbers: Anthropic's revenue has grown more than 10x annually over the past three years, and is now pacing at about $14 billion a year. Claude Code, Anthropic's AI coding agent, has doubled active users and revenue since the start of 2026. 📈 Stunning stat: The number of customers spending more than $100,000 annually has grown 7x in just the past year, Anthropic said. The company attributes the growth to its focus on enterprise customers, who are willing to pay more than everyday consumers. Go deeper.

[phkrause]

🏃‍♀️ AI outruns outside researchers

Illustration: Aïda Amer/Axios

AI is evolving faster than the systems designed to evaluate it. So lots of the scientific research you may read is already out of date by the time it's published, Axios' Herb Scribner reports. Why it matters: AI skeptics and critics will have to learn to keep up — or risk presenting misinformation themselves. 🤖 Case in point: A study from Oxford University this past week found that AI often gave wrong health advice, mostly due to how users asked questions. As pointed out by Kevin Roose, N.Y. Times tech columnist and podcaster, the study was based on users who worked with only three much older or little-known models — OpenAI's GPT-4o, Meta's Llama 3 and Cohere's Command R+. Roose said he's "begging academics to study AI capabilities using frontier models." Similarly, a study led by a Brown researcher found that using AI for therapy may breach ethical standards. But the study was done by prompting now-outdated LLMs. The bottom line: AI research can have a very short shelf life.

[phkrause]

🦠 AI's big biosecurity risk

Illustration: Sarah Grillo/Axios

Researchers from Johns Hopkins, Oxford, Stanford, Columbia and NYU are calling for guardrails on certain infectious disease datasets that could enable AI to design deadly viruses, Axios' Megan Morrone writes. Why it matters: Once high-risk biological data hits the open web, it can't be recalled — and regulation won't matter if the knowledge itself is already widely distributed. An international group of more than 100 researchers has endorsed a framework to govern certain biological data the same way we handle sensitive health records. Keep reading.   This doctor is training AI to do her job. And it’s a booming business Dr. Alice Chiao used to teach emergency medicine to students at Stanford University’s medical school. Now, she’s teaching artificial intelligence-powered chatbots to think, diagnose and prescribe like her. https://www.cnn.com/2026/02/17/business/ai-experts-training-jobs?

[phkrause]

AI put my kids to bed

Illustration: Aïda Amer/Axios

At 10:47 p.m. on a Tuesday, I was sitting on the floor outside my toddler's bedroom, so exhausted I could barely think. Then I did something that would have seemed absurd to me a year ago, Jamie Stockwell writes. I opened ChatGPT and typed: "I don't think I can put into words how much I love my children. They are truly my everything. But bedtime is breaking me." 😵‍💫 That night, in that moment of desperation, I didn't turn to parenting forums or to yet another expensive sleep-training program. I turned to AI, and it worked. 🤖 Why it matters: I'm far from alone in using AI for deeply personal problems. 1 in 6 people worldwide now use LLMs, according to research from the Microsoft AI Economy Institute. About 1 in 6 adults in the U.S. rely on AI daily for health advice, work scripts, emotional steadiness and logistics. 😴 The backstory: My 14-month-old was waking every 60 to 90 minutes, needing to be comforted. Separation anxiety turned bedtime into a marathon of tears. And my preschooler had mastered the bedtime filibuster — more stories, more snacks, more anything — pushing bedtime as late as 10:30. I was sleep-deprived, hopeless, and too tired to read another parenting book that promised answers ... but never quite fit our chaos. 🛟 What I got from a bot wasn't a checklist. It was a calm, steady presence coaching me through bedtime in real time, the way a seasoned sleep consultant might — if one lived in my house. I was skeptical. I described our nights in detail and expected generic advice: warm baths, healthy snacks, short books. We were already doing all of that — it wasn't working. Instead, ChatGPT responded: "I've got you. Let me give you a clear, compassionate starting framework, tailored to your style and to what's already worked for your family, while also honoring [your children's] temperament." 💤 What happened: What I got was a customized, seven-day sleep plan — and real-time help when everything fell apart at 3 a.m. 😬 Night 3: My toddler refused to lie down for 45 minutes. I asked what to do. ChatGPT coached me through staying calm and being consistent. "If he stands up 100 times, you guide him down 101 times. No emotion. No frustration. This teaches the boundary without fear." 🌪️ Night 5: My preschooler melted down, screaming for another book. Again, real-time guidance: "This… is the peak of the emotional storm. It feels awful, but it is not danger, and it is not a sign the plan is wrong." I was given an exact script for holding the boundary while staying present. 🎉 Night 7: Both kids were asleep by 8:30 p.m. I nearly cried with relief. Between the lines: My son started sleeping through the night for the first time in months. My daughter accepted our new routine: "one snack, two books, one song." 🌱 The big picture: Boundaries feel impossible when your child is screaming, and you're sleep-deprived and barely thinking clearly. This wasn't about outsourcing parenting. It was about getting the exact help I needed, in the moment I needed it, without judgment or shame. 🫶 The bottom line: My kids are sleeping. We have routines. I have my sanity back. And I got that help from generative AI. AI won't replace love or tenderness or the messy (but oh so rewarding!) work of raising small humans. But it can help parents feel calmer, more prepared and more capable along the way.

[phkrause]

🎬 Hollywood's AI clash with China

Illustration: Annelise Capossela/Axios

China's unrelenting efforts to catch the U.S. in AI may have claimed its first significant casualty — Hollywood, Axios' Madison Mills and Sara Fischer write. Why it matters: Technology good enough to scare even the most seasoned filmmakers is prompting a legal fight that's only the opening salvo in a broader war over intellectual property and market dominance. Chinese AI models that undercut U.S. rivals on price, speed and market share pose an existential threat to high-cost, high-risk industries like the film business. They also ship with fewer safety guardrails, especially around copyrighted material and likeness rights. 🔭 Zoom in: Seedance 2.0, ByteDance's new AI video model, generated a hyperrealistic clip of Tom Cruise and Brad Pitt fighting, prompting alarm from major Hollywood studios. Netflix, Paramount, Warner Bros. and Disney sent ByteDance cease-and-desists over the tool. Stunning stat: Chinese open-source models went from near-zero usage in mid-2024 to about a third of overall AI use by the end of 2025, according to OpenRouter.

[phkrause]

🤖 AI agent explosion

Illustration: Sarah Grillo/Axios

A manic new phase of the AI boom is sweeping through Silicon Valley, powered by autonomous "agents" capable of liquefying weeks of manual labor into minutes, Axios' Zachary Basu and Madison Mills write. Why it matters: For now, the frenzy is largely confined to software engineering. But inside that bubble, the shift feels seismic — deepening the gulf between AI builders and bystanders. "I've followed tech for 25 years, and I've never felt a larger gap between the ~1 million people using Codex/Claude and the rest of humanity," tweeted James Wang, director of product marketing at Cerebras — a chipmaking startup and Nvidia rival. 🖼️ The big picture: Anthropic CEO Dario Amodei recently described the current state of software engineering as the "centaur phase" — a reference to the half-human, half-horse creature of Greek mythology. Just as a chess player aided by a computer could once beat any standalone machine, an engineer paired with an AI agent may now be the most powerful unit in tech. Amodei argues that this hybrid phase may be "very brief" — perhaps only a few years, before AI systems can independently outpace even the best human-led teams. Zoom in: Major AI labs have spent the past year pitching "agentic workflows" as the industry's next frontier. That vision snapped into focus last month with the explosive rise of OpenClaw. Unlike chatbots that live in a browser or an app, OpenClaw gives agents "hands" on a user's local machine — letting them autonomously manage files, run terminal commands and message teammates. OpenClaw's founder, Austrian developer Peter Steinberger, was poached by OpenAI to lead its "personal agents" division. Data: Anthropic. Chart: Axios Visuals 🥊 Reality check: Meta and other tech firms have restricted or banned OpenClaw over fears that giving AI agents access to corporate systems could expose companies to malware, data leaks and manipulation.

[phkrause]

Get maximum chatbot value

Illustration: Brendan Lynch/Axios

If you're still using your chatbot like it's Google, stop. Stop it right now, Axios' Megan Morrone writes. Why it matters: Generative AI is fundamentally different — and far more useful — when you treat it like a collaborator and check its work. 🧑‍🏫 Think of yourself like a coach and the AI like the best player on your team. Then run the play. You're not alone if you're leaving value on the table. 🧰 Anthropic put out an AI Fluency Index today that showed people aren't using its Claude tools to their full advantage. The company assessed "AI fluency" using a framework developed with professors Rick Dakan and Joseph Feller. 📉 Stunning stat: "In only 30% of conversations do users tell Claude how they'd like it to interact with them," per the report. 🔁 Train the bot with the old adage: "Help me help you." The big picture: Generative AI keeps getting smarter and producing magical-seeming outputs, from code to legal briefs to possible medical diagnoses. Reality check: The more polished the output, the less users question it, the index shows. 💡 Some best practices based on the report: 1. 💬 Keep prompting: Treat the first response as a draft. Iterate, ask follow-ups, push back and refine. Ongoing back-and-forth is the strongest marker of real fluency. Example: "You got that wrong. Try again" or "Why did you give that answer?" 2. 🔍 Be skeptical of polished answers: When output looks great, pause. Check the facts, probe the reasoning and ask what might be missing. Don't lower your guard and drop your critical thinking. Example: Try your prompt with different chatbots and see if you get the same answer. Or ask: "Fact-check yourself and show me your sources." 3. 🛠️ Set the terms: Don't just prompt — instruct. Tell the AI to flag uncertainty or challenge your assumptions. Being explicit up front changes the dynamic. Example: Use global instructions and custom settings that most tools offer so you don't need to rewrite the same rules every time. The bottom line: AI fluency is core to your experience with the chatbot, no matter which one you pick. It's not always them; it's you.

[phkrause]

AI's biggest threats

Illustration: Sarah Grillo/Axios

If AI were a politician, it'd be headed for a landslide defeat, Jim VandeHei and Mike Allen write in a "Behind the Curtain" column. Defeated by Democrats ... Republicans ... and independents. Why it matters: In the Trump administration, Silicon Valley, and select AI-obsessed homes or businesses, people are euphoric about the fast rise of generative AI tools. These groups see a coming utopia. Almost everywhere else, the vast majority are indifferent, pessimistic — even downright dystopian. The politics are shifting so fast against AI that Democratic governors who championed it are in fast retreat. 🖼️ The big picture: It's almost certain to get worse for AI. Here's why: Every major AI company is racing frantically to pump out new, more human-like models and then boast about their awesome capabilities. Every advancement or boast likely causes an equal and opposite reaction from voters. They get more anxious. The AI companies are pouring money into politics, but it's mostly to thwart regulation they think could slow AI, not polish its image. So AI's image is shaped largely by critics or background noise. This weekend at the National Governors Association's annual meeting in Washington, Mike talked with governors from states big and small, red and blue. Most said voters not only fear AI's effect on kids and jobs, but they also find it creepy. "Many people think AI is either a science fiction movie or something that is going to take their jobs," Maryland Gov. Wes Moore (D), vice chair of the NGA, told us. "Government hasn't done a good job helping people separate fact from fiction. When people think about AI, they need to move beyond what they saw in a Will Smith movie." ("I, Robot" trailer) Indiana Gov. Mike Braun (R) is an optimist about AI as a force for good, especially in health care. But he told us many of the Hoosiers he talks to are worried about what China or North Korea might do. "When you get off to a bad start with an image, that's tough to fix," he said. Poll after poll puts numbers to this rising concern: 58% of Americans don't trust AI much or at all. 63% say AI will decrease the number of jobs in the U.S., according to an Economist/YouGov poll out last week. In a separate YouGov survey out in December, 77% of Americans were concerned AI could pose a threat to humanity. It's one thing to fear higher taxes. It's another thing to worry about the existence of your species! 79% of Americans don't trust companies to use AI responsibly, a Bentley-Gallup survey found. AI has more than a branding problem for a new product. After watching the effects of social media on kids and society, most seem to assume AI will just be worse. 👶 Threat No. 1: Kids. Nothing unites voters quite like fear about what AI is doing to children. This is the single most politically potent dimension of the backlash. Among women 50+ — high-turnout voters who'll play a pivotal role in midterms — 90% are concerned about the lack of national AI standards to protect kids, with 70% very concerned, according to Fabrizio Ward data. 💼 Threat No. 2: Jobs. The fear of displacement is deep, and the class divide is widening. Only 7% believe AI will increase jobs. That's statistically unchanged from last fall — meaning rising AI hype and use have done nothing to ease the fears. The more people read about super-workers using AI to 10x their output, the more scared it makes many others. 3 in 5 U.S. employees aren't currently using AI at work (Google/Ipsos last week). 😬 Threat No. 3: The creep factor. Beyond jobs and kids, there's a visceral, almost existential unease. That's one of the big takeaways from our conversations with governors. It's not just that people worry about AI taking jobs, or their power bills rising because of data centers. They also have real issues with the technology itself: People associate AI with surveillance, fakery and a loss of control. They worry about AI starting wars and tech ruling us. 🔌 Threat No. 4: Energy and land wars. AI's backlash is no longer just about screens and software. It's now about the physical infrastructure being built to power it — and the electricity bills landing in people's mailboxes. This is the most urgent political topic locally, where rising energy and land prices are stirring a massive backlash to data centers that run the size of multiple football fields and suck up a lot of power. AI isn't a top-tier issue — yet. Affordability and safety are dominating midterm campaigns. But it's climbing as AI power and hype rise. Interestingly, there's an emerging AI divide in both parties. You see anti-AI socialists like Bernie Sanders versus pro-AI leaders like Pennsylvania Gov. Josh Shapiro, though even Shapiro's stance is moderating as the winds change. And you see anti-AI populists like Steve Bannon versus pro-AI leaders like Trump, the entire White House and most of the GOP Congress. The bottom line: Most voters want to go slow on AI, don't trust business, and fear the technology could erode creative thinking and threaten humanity. Neither party is trusted to handle it.

[phkrause]

🪖 Grok enters classified AI race

Elon Musk's artificial intelligence company xAI has signed an agreement to allow the military to use its model, Grok, in classified systems, Axios' Dave Lawler and Maria Curi report.

• Why it matters: Until now, Anthropic's Claude has been the only model cleared for the systems used in the military's most sensitive intelligence, weapons development and battlefield operations.

Keep reading.

[phkrause]

Race to catch Claude

Illustration: Sarah Grillo/Axios

Supremacy can be fleeting in the AI race. But two months into 2026, Anthropic's Claude is upending U.S. national security, roiling financial markets and redefining how startups are built, Axios' Zachary Basu reports. Why it matters: The company is in the middle of the most important fight of the era — how much power to give AI in the face of threats, real and virtual. Anthropic said this week it's loosening the core safety pledge that defined the company, acknowledging that going it alone on restrictions won't work when rivals face none. The remarkable reversal came the same day the Pentagon threatened to effectively kick Claude out of government in a fight over its appropriate military uses. 🔭 The big picture: Two years ago, Anthropic was virtually unknown outside of San Francisco. Today, the startup is valued at $380 billion — raising $30 billion this month from some of the biggest financial and tech investors in America. Users consistently rank Claude above rivals for complex reasoning, nuanced writing and reliability. 🦾 State of play: Rivals are scrambling to catch up. OpenAI is expected, as soon as today, to release ChatGPT 5.3, or "Garlic" — the product of CEO Sam Altman's "code red" directive in December. But the biggest wild card is China, where the upcoming release of DeepSeek's V4 model threatens to reignite a U.S. market panic that wiped $1 trillion from tech stocks last January. Zoom out: For now, Claude has established its dominance across three engines of American power. 1. In Washington, Anthropic is locked in a high-stakes dispute with the Pentagon over whether Claude can be used for mass surveillance of Americans and for lethal weapon systems that don't require human involvement. Defense Secretary Pete Hegseth has given Anthropic CEO Dario Amodei until Friday to loosen Claude's military guardrails or face a potential "supply chain risk" designation. If that happens, anyone doing business with the Pentagon would be required to certify they don't use Claude — a potentially massive disruption, given it's already used by eight of the 10 largest U.S. companies. 2. On Wall Street, new releases by Anthropic have triggered five separate stock market gyrations in four weeks — a phenomenon traders have dubbed the "SaaSpocalypse." Feb. 3: Cowork legal plugins wipe out $285 billion in market value. Thomson Reuters plunges nearly 16% — its worst single day on record. LegalZoom craters 20%. FactSet drops more than 10%. Feb. 6: Claude Opus 4.6 launches and financial data stocks bleed again. The Nasdaq posts its worst two-day tumble since April. Feb. 20: Claude Code Security hits cybersecurity. CrowdStrike down 8%. Cloudflare down 8%. JFrog down 25%. Feb. 23: A Claude blog post about automating legacy bank code sends IBM to its worst single day since October 2000 — $31 billion gone by the closing bell. Feb. 24: Anthropic launches job-specific tools. First-wave victims — FactSet, DocuSign and Thomson Reuters — all rally after revealing new partnerships with Claude. 3. In Silicon Valley, Claude Code has become an obsession among venture capitalists and engineers who see it as the foundation of a new era of AI-native and agentic startups. Engineers describe Claude Code as the first tool that genuinely compresses development timelines from weeks to hours, allowing small teams to ship what once required entire departments.

[phkrause]

🤖 Altman: OpenAI has red lines, too

Photo illustration: Sarah Grillo/Axios. Photo: Kyle Grillot/Bloomberg via Getty Images OpenAI CEO Sam Altman says he'll draw the same red lines that sparked the high-stakes fight between rival Anthropic and the Pentagon, Axios' Maria Curi and Dave Lawler report. 📝 Altman said in a memo to staffers obtained by Axios: "We have long believed that AI should not be used for mass surveillance [which the Pentagon says is already illegal] or autonomous lethal weapons, and that humans should remain in the loop for high-stakes automated decisions. These are our main red lines." 🪖 Yes, but: Altman made clear that he still wants to strike a deal with the Pentagon to allow OpenAI's ChatGPT to be used in sensitive military contexts. 🏛️ Key Senate defense leaders are privately pressing Anthropic and the Pentagon to resolve their dispute, Maria and Ashley Gold scoop.

[phkrause]

Your AI glossary

Illustration: Natalie Peeples/Axios

AI is moving fast, and a wave of hot takes is fueling doomerism and amplifying fears about what tech could do to jobs, power and society, Axios' Madison Mills writes. Why it matters: When something scares me, I like to focus on what I can control. In this case, that's my own understanding of AI, as a reporter covering it. Here's a rundown of the AI power players and the shared language shaping the conversation. ⚒️ The heavyweights: Who's building what Several AI labs dominate the U.S. business and investor conversation. Here are the key firms, in alphabetical order: Anthropic: The maker of Claude. The CEO, Dario Amodei, has taken a safety-focused, business-first approach, launching tools like Claude Code and Cowork to attract business contracts. Google DeepMind: The maker of Gemini. CEO Demis Hassabis is a Nobel Prize-winning scientist focused on fueling research with AI. Google already has a strong business customer base, which could translate into future revenue for Gemini. Meta: The maker of Llama. CEO Mark Zuckerberg has positioned Meta as a major competitor to OpenAI and Google. The company releases so-called open models while weaving AI into Facebook, Instagram and WhatsApp. OpenAI: The maker of ChatGPT. CEO Sam Altman is focused on dominating the AI race. Altman's business model currently includes enterprise subscriptions and offerings like Codex, its coding tool. OpenAI is also beta testing ads as a potential future revenue stream. 💬 The buzzwords: What insiders are saying This is the shorthand used by executives and investors to describe how AI is getting more capable — and more independent. Vibe coding: Using AI to generate code from high-level prompts (aka, vibes). With vibe coding, a chatbot can build an app or website mostly by itself, but a human must still debug and refine it. Agent swarms: A "swarm" is a group of specialized AI agents working together to solve a complex problem. In the case of "agentic AI," AI-powered systems act autonomously to accomplish a task without consistent direction from a human. Recursive learning: When AI teaches itself, using its own outputs to inform its next version, potentially creating a feedback loop of rapid improvement without needing human-generated data. While recent models from major AI labs have helped train themselves, they are mostly still trained using human-created data. Human in the loop: This means what it sounds like: Keeping a person involved to review, approve or intervene, especially when AI systems act autonomously. Model Context Protocol (MCP): An open-source framework founded by Anthropic that lets a model securely connect and interact with other apps or data systems. This allows an AI model to talk to Excel or PowerPoint, executing tasks autonomously. METR Curve: Derived from the nonprofit METR (Model Evaluation and Threat Research), this tracks how long it takes AI to autonomously complete tasks without human intervention. That baseline has been increasing exponentially, doubling in about 7 months on average, according to their research. Industry insiders rely on this as a means of sussing out AI's progress. HALO: For markets nerds, it's an acronym for "heavy assets low obsolescence." The term is used to refer to the HALO effect around stocks or assets that are tangible (aka, not replaceable by AI). After years of a tech-driven rally, real-world stuff is cool again for investors. Gold is a recent example, up over 23% year-to-date. 🌟 The "North Stars": Where AI is heading AGI (artificial general intelligence): This is the point where an AI can perform any intellectual task a human can do. AI CEOs ranging from OpenAI's Altman to Google DeepMind's Hassabis increasingly hint that we are within a few years of this milestone. The Singularity: A hypothetical point in the future where technological growth accelerates beyond human control and becomes irreversible. The bottom line: The language of AI is still being written, but understanding it is one thing you can actually control.