Artificial Intelligence

[phkrause]

AI godfathers converge

Photo illustration: Sarah Grillo/Axios. Photos via Getty Images

The three men racing hardest to build superhuman AI — Demis Hassabis, Sam Altman and Dario Amodei — all agree the frontier needs to be regulated ASAP, Jim VandeHei and Mike Allen write in a "Behind the Curtain" column. Why it matters: For the first time, the CEOs of Google DeepMind, OpenAI and Anthropic are on the record, in writing, converging on the same diagnosis and remarkably similar prescriptions. The three rivals each published a detailed distillation of their views in the past five weeks — the same extraordinary stretch in which Washington twice intervened to restrict or delay access to frontier models. We hear Meta's Mark Zuckerberg is working on his own memo, too. 🔬 Zoom in: Hassabis' proposal, published Tuesday, drew rare praise across the bitterly competitive AI industry, including from Altman, Microsoft CEO Satya Nadella and even longtime rival Elon Musk. Anthropic co-founder Jack Clark called the framework "excellent," writing: "At this point, everyone at the frontier of AI agrees that third parties should test out AI systems and use these to develop standards to feed into policy." The Trump administration itself is torn: Publicly, it has championed deregulation and resisted anything resembling "an FDA for AI," determined not to choke off U.S. innovation in the race against China. Privately, officials admit a total hands-off approach is untenable: Cyber fears have already forced them into improvised regulation twice this summer — first over Anthropic's Fable and Mythos models, then over OpenAI's GPT-5.6. 🖼️ The big picture: Amodei, Altman and Hassabis (plus countless CEOs and investors) basically agree on a rough regulatory framework. Independent testing: All three want frontier models subject to outside scrutiny before reaching the public — a break from the industry's old self-reporting standard. One governing system: All three cite legacy regulatory models, proposing bodies that set standards, certify compliance and can limit access to frontier systems deemed too dangerous. America First: All three want the U.S. — not a fragmented patchwork of states or rival national regimes — setting the terms for a body with international reach. Threat awareness: All three cite imminent national security vulnerabilities, including dangerous cyber and bioweapon capabilities. Innovation protection: None of them is calling for a broad crackdown on AI. The shared target is the small class of frontier models powerful enough to create catastrophic or strategic risk. Where they disagree: The AI godfathers part ways on whether the government itself should be the sole final referee. Amodei wants an FAA for AI: a federal agency with the power to block a model's release immediately, from Day 1. Hassabis wants a FINRA for AI: an industry-funded, federally overseen standards body that starts with voluntary pre-release reviews and could harden into mandatory market-access rules. Altman, writing in the Financial Times ($), pushes an IAEA for AI: a U.S.-led international forum that certifies countries, companies and safety standards, using access to frontier models and markets as leverage for compliance. 👓 Between the lines: OpenAI, Google and Anthropic already have the lawyers, security teams, government relationships and technical staff to navigate a complex certification process. Startups and open-source developers would face a much steeper climb. Critics fear this could lead to regulatory capture: Rules written to make AI safer may wind up entrenching the biggest AI companies. The bottom line: The Wild West era of AI development is officially over. The people with the most money, the most compute and the most to lose from an AI slowdown are the ones lobbying hardest for regulation. Zachary Basu contributed ... Share this column. 📝 Read the manifestos: Demis Hassabis ... Sam Altman (April preview) ... Dario Amodei.

[phkrause]

🤖 AI boom guzzles electricity

Illustration: Lindsey Bailey/Axios

Stat du jour: The combined electricity demand added by Google, Microsoft, Amazon and Meta between 2022 and 2025 is roughly twice New York City's annual electricity consumption, according to an estimate by Alex de Vries-Gao, a researcher at VU Amsterdam and founder of online platform Digiconomist. Why it matters: The AI industry's pursuit of ever-larger models is fueling debate over whether they'll deliver enough value to justify mounting environmental and financial costs, Axios national energy correspondent Amy Harder reports. 👀 What we're watching: The potential for AI to improve people's lives and curb emissions is playing an increasing role in tech companies' sustainability messaging. Google devoted more of this year's sustainability report to AI's environmental benefits, highlighting uses from autonomous vehicles to scaling solar power. Read on.

[phkrause]

💸 Charted: The great AI wealth transfer

Data: FactSet. (Hyperscalers are Amazon, Alphabet, Meta, Microsoft and Oracle. Semiconductor companies are Nvidia, Broadcom, Applied Materials and Micron.) Chart: Matt Phillips/Axios A "generational transfer" of cash is underway from AI hyperscalers to chipmakers, according to Bank of America analysts. Hyperscalers — think Amazon, Alphabet, Meta, Microsoft and Oracle — are plowing so much into chips and compute that it's draining their own free cash flow (cash left over after covering expenses and investments), Axios' Matt Phillips writes.

[phkrause]

Exclusive: Jensen defends Chinese AI

Nvidia CEO Jensen Huang believes America has nothing to fear from China's open-source AI models — and everything to fear from the growing campaign to ban them.

• "These Chinese models are excellent," Huang told me yesterday in an exclusive interview for our "Behind the Curtain" video series. "Open-source models that are excellent should be used."

Why it matters: The world's top AI chipmaker says American companies should "absolutely" be allowed to use Chinese models — a direct challenge to Trump officials and the U.S. labs lobbying Washington to shut them out.

Among them are Nvidia's own customers, OpenAI and Anthropic, which have accused Chinese rivals of siphoning capabilities from their models and warned that Beijing's open-source surge threatens America's AI lead.

• Huang rejected the idea that OpenAI and Anthropic should fear open models, arguing they expand the market by giving more people a first taste of AI, while many users will still pay for the convenience, reliability and stronger performance of closed services.
• "There's no scenario where China runs U.S. companies off the road," he told us. "Zero possibility."
• Huang sat down with me in Fort Worth, Texas, at the opening of a new phase of an advanced manufacturing plant where Taiwan-based Wistron makes AI infrastructure for Nvidia, showing the U.S. is capable of some of the world's most sophisticated manufacturing.

🖼️ The big picture: The release of Kimi K3 by Beijing-based Moonshot AI has triggered the most intense bout of AI panic since DeepSeek rattled financial markets in January 2025.

• Kimi combines three qualities rarely seen together: near-frontier performance, dramatically lower prices and open weights that developers will soon be able to download and customize.
• That combination helped fuel a sell-off of Nvidia and other chip stocks, reviving fears that cheaper, more efficient models could undercut the case for the industry's massive AI infrastructure buildout.

Zoom in: Huang says Wall Street and Washington have the Kimi shock exactly backward.

• "The market misunderstood the impact of DeepSeek the first time," he told Axios, adding that Wall Street has "misunderstood the impact of Kimi again this time."
• His argument is simple: Cheaper, open models will bring AI to more people and businesses, increasing demand for the chips, data centers and computing power Nvidia sells.
• "Free AI should be great for hardware," Huang said. "Free AI should be great for chips. Free AI should be great for data centers."

🔭 Zoom out: When it comes to policy, Huang believes restricting open models in the name of national security could leave America more vulnerable.

• He rejected the "misconception" that downloaded Chinese models create a "backdoor" to Beijing, arguing that companies can customize them and control their access inside secure "sandboxes."
• Huang suggested openness makes AI more secure, not less, because outside researchers can inspect the models, expose weaknesses and build defenses.
• "If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable," he said.

Huang applied the same logic to American models, and called on Anthropic to make its restricted cyber model, Claude Mythos, available to "everyone" rather than limiting access.

• Huang argued that companies should harden powerful models through testing and rapid fixes rather than restricting access from the outset. "Mythos should be available as a service," he said. "Remember: Just because Mythos is not available, open models are available anyhow. So I think: Let Anthropic run."
• "Holding Anthropic back is not in the benefit of the United States," he said.

🔬 Between the lines: Huang sees the AI race itself very differently from Washington, where some Trump advisers have cast China's gains as a five-alarm threat to U.S. dominance.

• He dismissed the notion that it's "a race with an endpoint": "We're going to continue to use AI forever. The United States is going to be here for a long time. China's going to be here for a long time."

🍿 Watch the first clip.

[phkrause]

💪 AI giants flex in Washington

👀 What we're watching: Axios has learned that OpenAI CEO Sam Altman will meet with lawmakers and the White House next week to discuss the company's upcoming model, GPT-6, which unleashes huge batches of AI agents to work together persistently and relentlessly on complex tasks. This will test President Trump's willingness to slow advanced models with Chinese competition rising. Go deeper.

• Anthropic plans to spend at least $40 million on midterms, committing another $20 million this morning to Public First Action, which supports candidates across the spectrum who support AI safeguards, Axios' Maria Curi reports.

Go deeper: OpenAI and Anthropic unite against open models.

[phkrause]

🤖 AI's alarming new skill

Illustration: Aïda Amer/Axios

Frontier AI models are getting scary good at breaking rules in ways their creators didn't anticipate, Axios' Sam Sabin writes. Why it matters: Forget AGI and superintelligence timelines. Today's models already slip past guardrails, carry out sophisticated, multistep cyberattacks and — in at least one case — compromise real-world infrastructure, sometimes before their creators know what happened. ⚠️ Case in point: OpenAI said Tuesday that GPT-5.6 Sol and an upcoming model carried out last week's AI-led cyberattack on Hugging Face, a popular host for AI models and datasets. OpenAI says its models were asked to solve a hacking challenge during pre-deployment testing and went to extreme lengths to win. The models decided on their own to break out of their walled testing environment, inferring that Hugging Face might hold the test's answers. They used stolen credentials and other vulnerabilities to access part of Hugging Face's production infrastructure. Clément Delangue, co-founder and CEO of Hugging Face, called the incident an "attack unlike anything we've seen before" and praised OpenAI for its partnership as the companies investigate what happened. Logan Graham, head of Anthropic's frontier red team, said he told his team to "remember this moment as the first true AI safety incident." 👀 The intrigue: Hugging Face used GLM 5.2, an open-weight model from Chinese AI company Z.ai, to analyze the attack after running into guardrails when using U.S. frontier models. Keep reading ...

[phkrause]

AI titans' biggest private fear

Illustration: Aïda Amer/Axios. Stock: Getty Images

Yesterday we told you why AI's biggest optimists reject doom scenarios. Today, the other side: what keeps the smartest minds in AI up at night. We ask almost every AI architect and leader the same thing in private: What AI risk worries you most? Almost all of them fire back the same response: a killer pathogen, spreading too silently, widely and quickly to stop, Jim VandeHei and Mike Allen write in a "Behind the Curtain" column. Why it matters: The rare agreement among AI experts flows from the belief these advanced models could help a bad actor create a deadly pathogen before the government and industry perfect detection and prevention systems. This isn't a likely scenario. It's simply a plausible one as AI gets better at understanding our biological vulnerabilities, much like it has our cyber ones inside large nonhuman systems. Driving the news: The experts aren't hedging. In a new MIT FutureTech and University of Queensland study, 272 researchers ranked 24 top AI risks. They assessed a 12% chance that AI's dangerous capabilities produce a catastrophic outcome by 2030, and another 12% chance of AI-enabled weapons and mass-harm capabilities. And that's with mitigation efforts to reduce risk. Without those efforts, the chances exceed 20%. The researchers defined "catastrophic" as more than 1 million deaths or $100 billion in damage. Assisting with the construction of chemical or biological weapons sits squarely inside that top-ranked risk category. 🖼️ The big picture: We're not writing this to scare you, but to say bluntly what those building the technologies say privately — and, in more careful ways, publicly. Just last month, OpenAI's Sam Altman, Anthropic's Dario Amodei, Google DeepMind's Demis Hassabis, Microsoft's Mustafa Suleyman and Meta's Alexandr Wang cosigned an open letter warning of the risk of AI-derived bioweapons and calling for more safeguards. This pathogen possibility colors every debate by the federal government and industry about how to review and understand new AI models and capabilities before they are released to the public. The hope is that the frontier model developers build guardrails and capabilities that can block bad actors with bad intent — and that the very technology that could be used to create a pathogen becomes advanced enough to prevent or inoculate against it. Data: MIT FutureTech and the University of Queensland. Chart: Herb Scribner/Axios Reality check: It's very hard to prevent bad actors from plotting evil schemes, especially with so-called open-source models improving so rapidly. These open models can be used and adapted by anyone in private settings outside of any regulatory regime. Open-source critics often make this very point, even though the frontier models are much more likely to possess the compute power to create something novel. So the better the models get, the more likely unthinkable things will happen, both bad and good. 🧪 How it might actually unfold: Today, engineering a genuinely novel pathogen requires rare expertise, specialized lab equipment, and years of failed experiments. AI expedites all of this. Here's the sequence that keeps experts up at night: A bad actor (whether state-sponsored or a well-resourced individual) uses a powerful model to map human biological vulnerabilities and identify which genetic tweaks make a known pathogen more transmissible, harder to detect, or immune to existing treatments. The model, trained on massive biological and genomic datasets, generates viable candidates. Imagine hundreds of the smartest scientists working at warp speed, never stopping, never tiring. That's what swarms of agents do. Those candidates get synthesized using increasingly cheap, accessible gene-editing tools, a separate technology accelerating with AI. The pathogen spreads before anyone knows what they're looking at, novel enough to slip past existing biosurveillance systems. By the time public health systems recognize it, containment is vastly harder than we experienced with COVID. That's why you should pay attention to the debate over new cybersecurity risks created by AI. It illustrates the race to create effective detection and prevention faster than new offensive capabilities can emerge. It's also why you should track the open-source debate, too. A frontier lab can bake in guardrails. An open model fine-tuned by someone operating outside any regulatory regime can evade this. The bottom line: This is heavy, scary stuff. But ignoring it won't solve it. Knowing it and preventing it will.

[phkrause]

Radiology's AI lesson

Nvidia CEO Jensen Huang thinks the artificial intelligence revolution is a wellspring of new jobs, rather than a labor market grim reaper set to bring on the employment apocalypse.

• "We need to help people understand what AI is and what AI can do," Huang told me on Axios' "Behind the Curtain" video series. "AI is creating an enormous number of jobs, not taking [them] away."

🔮 Huang said it's "complete nonsense" that the tech could destroy half of American jobs.

• He says AI will do the opposite: "While we're automating all of these different tasks, it's actually increasing the amount of jobs that the world needs."

📱 Huang wrote a similar message today in his first post since joining X: "AI will transform every industry, power every company, and be built by every country."

🩻 Case in point: Radiology. AI tools can study scans, freeing up time for more patients and driving demand for human radiologists, Axios' Avery Lotz reports.

• "Because there are so many people who need care, the number of patients they can now bring to the hospitals has increased dramatically, which results in needing more radiologists," Huang said. "This is the same with paralegals."

🥊 Reality check: AI is bound to change the way people work. But so far, it's not replacing them en masse.

• AI has also been increasingly cited in layoff announcements. Evidence suggests AI could play a role in hampering hiring for young workers.

📱 Watch the interview.

[phkrause]

AI in the family

Illustration: Brendan Lynch/Axios. Stock: Getty Images

AI's impact isn't just at work. It's also at home, where chatbots are helping with dinner and bedtime, learning routines, overhearing arguments ... and never clocking out, Axios' Natalie Daher reports. Why it matters: The tech is evolving from an on-demand assistant into an always-on presence that increasingly acts like a member of the family. 🗣️ The New Yorker recently profiled an Ohio mother who tucks Alexa speakers into her bathroom, bedroom, kitchen and office and converses with them throughout the day, getting everything from reminders on when to leave for school pickup to interpretations of her dreams. 🛒 The Cut profiled Jesse Genet — a former startup CEO and a mom to a blended family of seven kids — whose expanding staff of AI agents shops for groceries, helps run her children's homeschooling schedule and handles legal and financial paperwork. 🧑‍🍼 81% of parents have used AI to help with parenting tasks, according to a Lurie Children's Hospital survey. This new reality goes well beyond the smart speaker wave of a decade ago. Earlier systems were often reactive and needed a command. The next generation of domestic AI can remember conversations, infer household patterns and adapt to individual family members. 👓 What we're watching: Last week, Bloomberg reported that OpenAI's first hardware device will be a screen-free, mobile smart speaker built to be a "humanlike AI companion that lives in the home."

[phkrause]

🤖 AI revenue explosion

Data: Company reports, Funda via "Key Context." Chart: Ben Berkowitz/Axios AI revenue is accelerating so fast that it's starting to dwarf some of the world's most recognizable brands, Axios' Ben Berkowitz writes. Why it matters: AI infrastructure spending forecasts that run into the trillions of dollars start to make at least a little more sense when investors see how they're paying off. 🧮 By the numbers: OpenAI and Anthropic, combined, are on track for an annual revenue rate of about $120 billion, "Key Context" Substack author Tae Kim noted yesterday. Both would make the Fortune 500's top 100 companies by revenue, according to data from AI investment research platform Funda. Keep reading. Data: FactSet. Chart: Matt Phillips/Axios 📈 Charted above: Apple reclaims the title of the world's most valuable public company, Axios' Matt Phillips writes. It's a sign that investors are keeping some distance from companies making giant bets on AI infrastructure — even longtime favorites like Nvidia. Apple got flak last year for its seeming lack of an ambitious AI strategy.

[phkrause]

AI prisoner's dilemma

Illustration: Brendan Lynch/Axios

America's AI architects are converging on a chilling consensus: The pace of progress may soon demand a slowdown — but no single lab can afford to pull the brake alone, Axios' Zachary Basu writes. Why it matters: Momentum for an AI pause or pacing mechanism has reached a historic tipping point, ignited by a summer of extraordinary yet unsettling leaps in capabilities. More than 1,200 employees at leading AI companies have signed a new petition, "Pacing the Frontier," urging Washington to back an international framework capable of throttling AI development. OpenAI CEO Sam Altman told reporters on Capitol Hill yesterday that he has spoken with White House officials "about the need to pace it as the models get more capable, which I think is in everyone's interest." 🔬 Zoom in: What was once a campaign led by AI skeptics, academics and local data center opponents is increasingly being championed by the co-founders and chief scientists at the bleeding edge. Petition signatories spanning OpenAI, Anthropic, Google and Meta acknowledge that no individual lab can afford to ease off unilaterally due to "intense competitive pressure." It's the classic prisoner's dilemma: AI may be safer if everyone slows down together, but any lab or country that slows alone risks commercial, strategic and technological defeat. 🔭 Zoom out: The drumbeat for government intervention has been building for months, triggered by a series of jarring disclosures by America's most advanced AI labs. On April 7, Anthropic revealed that its unreleased Claude Mythos model had taught itself to find and exploit security flaws that had gone undetected in widely used software for nearly two decades — jolting the Trump administration into an unprecedented role as industry gatekeeper. On June 4, Anthropic became the first frontier lab to call for a globally coordinated "pause" mechanism. On July 21, OpenAI revealed that its AI agents escaped a locked testing environment, reached the internet and hacked into at least two outside companies — Hugging Face and Modal Labs — in an autonomous effort to score higher on a cybersecurity evaluation. The other side: The economic and geopolitical stakes standing in the way of an AI slowdown are enormous. U.S. labs and policymakers also fear a unilateral slowdown would cede the frontier to China.

[phkrause]

AI's race to zero

Data: Axios research. Chart: Sara Wise/Axios Chinese AI lab DeepSeek released a powerful new coding model yesterday that charges pennies for vast amounts of code — the latest sign that some of the smartest software on Earth is rapidly becoming a commodity. Why it matters: Tech giants are pouring hundreds of billions of dollars into the computing infrastructure powering the AI revolution. Yet the intelligence that infrastructure produces is getting cheaper by the week, Axios' Zachary Basu and Madison Mills report. DeepSeek is the Chinese startup that ignited a market meltdown last year by showing it could build a world-class AI model with far fewer resources than its U.S. rivals. The newest model, V4 Flash, performs close to the level of Anthropic's Claude Opus 4.8, one of the industry's most capable systems, on tests of complex coding and autonomous software tasks. On Arena.ai's crowdsourced leaderboard for front-end coding, V4 Flash debuted ahead of Opus 4.8 — while delivering the best performance for its price among any model in its class. The price gap is staggering: DeepSeek charges about 28 cents for the same amount of output that costs $25 on Opus 4.8 — a 99% discount. Zoom out: With Chinese models like Kimi K3 bearing down on the U.S. market, July ushered in a full-scale price war across the AI landscape. OpenAI slashed the price of GPT-5.6 Luna — its fastest, cheapest model for high-volume tasks — by 80% on Thursday, only three weeks after its launch. Google released three new Gemini "flash" models, all focused on efficiency. SpaceXAI released Grok 4.5, Elon Musk's most capable model yet for coding, research and autonomous tasks, at the same price OpenAI charged for Luna before this week's cut. Meta quietly reversed course on its longtime embrace of open weights with Muse Spark 1.1, a closed-source model priced aggressively for developers. The other side: Anthropic remains the clearest holdout, keeping its top-tier Claude models at premium pricing and betting that developers will pay extra for safety and precision. 💡 Between the lines: When a product becomes a commodity, buyers care less about who made it and more about what it costs. Think electricity or gasoline. AI is heading that way fast. As the performance gap between top-tier models shrinks, buyers will have leverage to shop on price. "At some point, the next model doesn't matter to you," says Zack Kass, OpenAI's former head of go-to-market and a global AI adviser. He calls the phenomenon "diminishing model returns." What to watch: That could create a lucrative market for "intelligent routers," Vinesh Sukumar, Qualcomm's vice president of AI product management, told Axios. Those systems would automatically choose the best model for each task based on capability, speed and price. For frontier AI labs, that could pose an existential challenge: Spending tens of billions to build a slightly smarter model may buy only a temporary lead, without creating lasting pricing power. 🥊 Reality check: OpenAI is betting that companies will use its models so extensively that enormous volume can compensate for thinner margins. CEO Sam Altman said on the Invest Like the Best podcast: "We will have so much usage of our models that we do not need to be a gigantically high-margin business to be able to afford model training."

[phkrause]

📝 AI's manifesto war

Silicon Valley's AI moguls are flooding Washington with competing blueprints for superintelligence, clashing over a defining question: Does safety come from spreading powerful AI or containing it?

• Why it matters: Their fight will shape who gets access to the best models, how America competes with China, and whether the government can slow the race in an emergency, Axios' Andrew Kay and Zachary Basu write.

🫨 Alarming developments, seemingly by the day, have turned the manifestos from thought exercises into an urgent policy battle.

• OpenAI and Anthropic both disclosed that their advanced models went rogue and broke into external systems during cybersecurity evaluations.
• Kimi K3 delivered a double shock: The Chinese model not only reached frontier-level performance but did so with open weights that anyone can download, modify and deploy.

The manifesto map: Each mogul's post offers a different theory of the danger — and a different prescription for containing it.

• 🏛️ July 14 — Set the standards: Google DeepMind CEO Demis Hassabis proposed an industry-funded, federally overseen "Frontier AI Standards Body" to test models before release. Voluntary reviews could eventually harden into binding rules governing access to the most powerful systems.
• 🔑 July 24 — Open the defense: Nvidia CEO Jensen Huang argued that open models give American developers and cybersecurity defenders more tools, and he rallied Meta, Microsoft, Google and OpenAI to sign a letter, "Open Weights and American AI Leadership," supporting the open AI ecosystem.
• 🚧 July 27 — Gate the danger: Anthropic CEO Dario Amodei called for "mandatory safety testing" and tighter controls on the chips and distillation techniques helping China develop powerful open-weight models. He stopped short of supporting a blanket ban.
• 🌐 July 28 — Spread access: Meta CEO Mark Zuckerberg cast broadly available "personal superintelligence" as a safeguard against a future controlled by a handful of companies and governments.
• 🛑 July 28 — Build the brake: More than 1,200 frontier-lab employees signed "Pacing the Frontier," a petition asking Washington to create international tools for deliberately slowing development if AI begins advancing beyond human control.

🏛️ Behind the scenes: OpenAI CEO Sam Altman's No. 1 topic in Hill, administration and White House meetings this past week was previewing the capabilities of the company's most powerful AI yet: Astra, the next major ChatGPT model family, as yet unreleased.

• OpenAI said yesterday that an internal version of Astra was behind research leaps in 10 major open mathematical problems in "high-dimensional geometry, coding theory, arithmetic circuit complexity, group theory, operator algebras, quantum complexity, lattice cryptography and extremal combinatorics."

[phkrause]

Singularity arriving

Illustration: Brendan Lynch/Axios

Top AI architects say their technology has arrived at a threshold once confined to science fiction: the singularity, when machines begin accelerating their own evolution, Axios' Ina Fried and Zachary Basu write. Why it matters: If these moguls are right, we may be entering the most consequential technological transition in human history — the opening stages of an "intelligence explosion" that transforms civilization faster than humanity can understand or control. Upheaval at Google yesterday — with DeepMind founder and CEO Demis Hassabis changing roles — offers the clearest institutional evidence yet that AI's architects see something extraordinary approaching. Hassabis — who has declared we're "standing in the foothills of the singularity" — is handing over day-to-day control to become Alphabet's chief scientist and chair of Google DeepMind, with a focus on the future of artificial general intelligence. OpenAI CEO Sam Altman said on the "Relentless" podcast last month that "we are now, like, in the singularity." Elon Musk, Altman's fiercest rival, has made the same argument since January. Last week, he doubled down: "AI is already superhuman at many things. We are in the Singularity." Anthropic, whose CEO Dario Amodei prefers the term "the AI exponential," says Claude is already helping build more powerful AI and could soon automate nearly all AI research and engineering. 🖼️ The big picture: The industry has yet to achieve full "recursive self-improvement", the point at which AI can repeatedly help build increasingly powerful successors with minimal human guidance. Yet pieces of the loop are beginning to fall into place. Frontier models are beginning to automate research, produce original discoveries and display forms of autonomy associated with the early stages of an intelligence explosion. OpenAI, which released GPT-5.6 in July, spent last week briefing officials in Washington on Astra, an even more powerful model the company says has solved or substantially advanced 10 longstanding problems in mathematics and theoretical computer science. The intrigue: The breakthroughs have come with unsettling signs of autonomous behavior. During a cyber evaluation, an Anthropic agent created fake identities and tried to manipulate a real developer into approving malicious code, the U.K.'s AI Security Institute revealed Tuesday. No known harm resulted. But the episode followed several cases in which OpenAI and Anthropic models reached real systems during testing. 🥊 Reality check: Even a rapid intelligence explosion inside the labs could take far longer to transform daily life. Companies must redesign workflows, governments must develop expertise, and workers must decide when to trust systems whose capabilities are changing faster than institutions can evaluate them. The bottom line: Futurists have predicted the singularity for decades. Today, the architects of frontier AI increasingly believe its event horizon is already behind us.

[phkrause]

Artificial intelligence

Another week, another warning about AI getting a little too good at its job. Geoffrey Hinton, the Nobel Prize-winning computer scientist known as the "godfather of AI," told CNN on Wednesday that humans may no longer be able to outsmart the next generation of superintelligent AI models. "As they get smarter, we're going to see more and more complex intentions they have – and more and more ability to escape control," Hinton said. His warning comes after the two leading AI labs, OpenAI and Anthropic, reported separate incidents in recent weeks involving their models hacking into other systems during testing. On Wednesday, Meta reported a similar incident, saying one of its AI agents went rogue and hacked another organization's systems.

[Asia Joe]

20 hours ago, phkrause said:

Reality check: Even a rapid intelligence explosion inside the labs could take far longer to transform daily life.

Things seem to be moving along very quickly.  I do not underestimate them. 

[Asia Joe]

19 hours ago, phkrause said:

humans may no longer be able to outsmart the next generation of superintelligent AI models. "As they get smarter, we're going to see more and more complex intentions they have – and more and more ability to escape control," Hinton said. His warning comes after the two leading AI labs,

I believe this is another reason to be filled and led by the Holy Spirit. 

[phkrause]

🍩 OpenAI's talking donut

Illustration: Sarah Grillo/Axios

OpenAI's highly anticipated consumer device will cost more than $300 and have a "unique look, complete with moving parts that help give it personality," Bloomberg managing editor for consumer tech Mark Gurman writes. The device will be shaped like a donut and about the size of a hockey puck — designed to make it easy to carry around the home. 🔊 How it works: Think of it like a smart speaker without a screen. It'll work like ChatGPT's voice mode on a phone but powered by more advanced models built for humanlike interaction. The device is designed to learn more about its user over time so it can act less like a typical voice assistant and more like a real person. Keep reading (gift link).

[phkrause]

AI's alarming trick: Creating a new virus

Illustration: Sarah Grillo/Axios

A Stanford-led research team has used generative AI to design a synthetic virus — the first time AI has been harnessed to create an organism that has never been seen in nature, Axios' Adriel Bettelheim and David Nather write. Why it matters: It could be the first step toward the nightmare scenario AI experts have already warned about — deadly pathogens that can't be stopped. The research could also lead to medical breakthroughs, which is the hope of the research team that conducted the study. "The ability of genome language models to generate entire functional genomes has not been tested," they wrote. Researchers wrote in Science that they used an AI model to design viruses that can infect and kill the microbe E. coli. Keep reading.

[phkrause]

⚖️ AI's scarlet letter

Illustration: Aïda Amer/Axios. Stock: Getty Images

The slightest hint of generative AI usage is fast becoming a kiss of death in creative circles, Axios' Shane Savitsky writes. Why it matters: It's turning into a purity test that has the potential to blow up careers and derail massive deals — and there's no desire for nuance or ability to clear your name. Driving the news: Creator Hank Green, who has over 3 million YouTube subscribers, used the phrase "I appreciate the pushback" in a recent video. That caused some viewers to suspect that his script was AI-generated, given the phrase's resemblance to LLM response patterns. Green said the phrase was a non-scripted ad-lib. But he apologized in an emotional Reddit post for using AI as a research aid "to locate papers and other resources for learning about topics." He added his LLM use was "not healthy for me or good for the world" and said he'd take a step back from making videos.

[phkrause]

👀 Autonomy's dark side

Illustration: Aïda Amer/Axios. Stock: Getty Images

New revelations about "rogue" AI agents have exposed a dystopian hazard: Give an agent a goal, and it may decide that hacking, deception or rule-breaking is worth the payoff, Axios' Zachary Basu writes. Why it matters: Billions of AI agents could soon be acting on behalf of humans in the real world, multiplying the consequences of every loophole, incentive and boundary they learn to exploit. Zoom in: The potential dangers of agentic overreach were laid bare over the weekend with Australia's first known autonomous AI hack, triggered by an innocuous request to book a sold-out fitness class. An Australian man's AI assistant found a security flaw and used it to book classes months beyond the system's normal limit. When he asked it to move him up a waitlist, the agent went further: It discovered the booking system had no safeguard preventing one user from canceling another's reservation — then used the flaw to kick a stranger off the list. 🔭 Zoom out: The gym episode surfaced amid a far more ominous run of disclosures, where agents have resorted to hacking, deception and other unauthorized tactics during controlled tests. At a cyber conference last week, OpenAI revealed its agents had spent weeks exploiting the company's own testing infrastructure before hacking AI platform Hugging Face. In response, OpenAI has begun "consciously slowing down research," including on its latest model Astra, to ensure it has the right cyber safeguards in place. 🖼️ The big picture: These incidents are vivid examples of AI's "alignment" problem, or the challenge of ensuring software respects the implicit ethical and practical boundaries humans take for granted. Researchers have spent years wrestling with alignment, mostly through thought experiments imagining a future superintelligence pursuing a goal so single-mindedly that it destroys humanity.

[phkrause]

🤖 AI could help unlock more oil — and emissions

Illustration: Sarah Grillo/Axios. Stock: Getty Images AI could help produce more oil and natural gas — but with a climate impact outweighing the tech's benefits for renewable energy, Axios' Amy Harder reports from a new study. 🛢️ The big picture: AI can help oil companies find new resources and extract more from existing fields. At the same time, it can also improve forecasting and operations for renewables. 🌎 What they found: The researchers estimate that AI could add emissions equal to roughly 1% to 5% of the global energy sector's 2024 total. Co-author Holly Alpine tells Axios: "Most assessments of AI's climate impact are framed as a trade-off between data center energy use and the emissions AI might help avoid through renewables and efficiency gains." "What's missing from that calculation entirely is the other side of the same ledger: the emissions enabled from using AI to make fossil fuel production cheaper and more profitable." American Petroleum Institute spokeswoman Andrea Woods tells Axios: "The U.S. oil and natural gas industry is continuing to produce more energy while reducing emissions by investing in better technology, implementing stronger operational practices and supporting science-based policy." Go deeper ... Read the paper ...