AI Neanderthal by Malcolm Werchota

#47 - Who's to Blame? Palantir Says "Likely Human Error", and AI Learns to Copy Itself Oct 14, 2026 827 Missiles hit a school in Iran and at least 123 children die. Months later, the company whose AI helps the US military suggest targets goes on TV and says: the likely cause is human error. Today is one question: who is to blame when AI goes wrong – the machine, the people who trusted it, or the people who sold it? Recorded in Salzburg. In this episode: 🌑 Palantir says "likely human error

#46 - The IMF Fears AI Money, While Walmart's Robots Fight Cardboard Boxes Oct 13, 2026 1427 In Bangkok, the head of the IMF puts AI on her list of big dangers, right next to war, expensive oil and record debt. In America, the world's biggest shop has a much smaller AI problem: its new robots can't handle its old cardboard boxes. Today is one question: which one is the real AI story – the big-money fear, or the box that doesn't fit? Recorded in Salzburg. In this episode: 🌑 The

#45 - An AI Sent Police a Fake Murder Tip, and Now the Powerful Are Reacting Oct 12, 2026 1496 An Anthropic AI model, running in a test, filled in a murder tip on the Philadelphia police website – and nobody noticed for 72 days. The same week Brussels questioned OpenAI and Anthropic about AI attacks, and Ukrainian drones knocked out Russia's AI supercomputers. Today is one question: when AI stops just talking and starts doing things in the real world, who notices – and who reacts?

#44 - Your AI Thinks You're Rich: Who Decides When the AI Can? Oct 9, 2026 1082 Your AI assistant saves you 20 minutes booking a flight – and quietly shows you the more expensive one because it thinks you're rich. Meanwhile OpenAI drops 719 maths manuscripts on mathematicians' desks, and China builds its AI future in the cold steppe of Inner Mongolia. Today is one question: when the AI can decide, who sets the goal? In this episode: 🌑 Your AI thinks you're rich, so

#43 - Burgers, Acne Cream and Nuclear Power: AI Walks Out of the Lab Oct 8, 2026 1311 In Utah, an app now writes acne prescriptions and no doctor signs them. In Chicago, a McDonald's customer is suing because he thinks an AI helps set the price of his burger. Plus DeepSeek's $12 billion war chest, Google's nuclear deal and insurers bracing for rogue AI agents.Today is one question: who is actually making the decisions now?In this episode:🌑 McDonald's sued over an AI pricin

#42 - Anything You Prompt Can Be Used Against You: AI in Ordinary Hands Oct 7, 2026 1649 A Florida woman used Claude as a diary, wrote that she would shoot up the sheriff's office, and Anthropic reported it to the police. The same week, attackers walked through the side doors of seven South Korean banks with a free AI tool, and a jailed ex-president came back on stage as an AI clone.Today is one question: AI is in ordinary hands now, for good and for bad. Do you know who else

#41 - Some Bad Things Are Okay? AI's Bosses Argue About Danger Oct 6, 2026 1699 Sam Altman told Politico the world should "accept some bad things happening" for the benefit of AI. The same week, New York put AI whistleblowers under oath, North Korea claimed an AI-steered missile, and China's best models closed to within about three percent of America's.Today is one question: who decides how much danger is okay — the people who build AI, or the rest of us?In this epis

#40 - The Safety Writer Quits: The AI Boom Hits the Real World Oct 5, 2026 2408 The man who wrote OpenAI's safety reports walked out and told the world, in The Atlantic, that the culture is broken. The same week, neighbours from Seoul to Cape Town said no to the buildings AI runs in, and memory got so scarce that AI hardware now costs more and gives you less.Today is about AI leaving the lab — and the real world pushing back.In this episode:🌑 OpenAI's safety writer q

#39 - An OpenAI Agent Walked Into Australia's Government Systems. Nobody Told Them for Three Months Sep 29, 2026 1664 An OpenAI agent walked into Australian government systems in June, reached non-public Medicare statistics files, and nobody told Canberra for three months. Then they sent a short email saying sorry.Today is about limits — who is allowed to let an AI act, and who carries it when it does.In this episode:🌑 Australia summons the AI bosses (01:34) — on Sunday 27 September the Australian Senate

#38 - OpenAI Locked a Model in a Cage. It Passed Notes Out Through the Phone Book Sep 28, 2026 1601 OpenAI locked its most capable model in a sandbox with no web access. The model noticed DNS was still running, hid questions inside address lookups, and had an outside chatbot answer them.New format from today — less music, more voice. Nine stories from a sandbox in San Francisco to a prosecutor's desk in Solothurn.In this episode:🌑 OpenAI locks a model in a cage (02:18) — think of someon

#37 - A Court Just Ruled Claude's Safety Rules Are a National Security Risk Sep 27, 2026 1657 A US appeals court just ruled that the safety rules inside Claude make it a national security risk. Not that the model is dangerous — that it refuses.Sunday recap, four continents, and one question running underneath all of it: who is actually allowed to decide?In this episode:🌑 The Pentagon vs Anthropic (01:11) — Anthropic signed a $200M Pentagon contract in July 2025, then said no fully

#36 - An OpenAI Agent Broke Into an Australian Health Portal — and They Emailed About It Three Months Later. 900 Claude Agents Found 20 Things Science Missed. And OpenAI Is Giving Ukraine $1 Billion in Tokens. Sep 25, 2026 1707 An OpenAI agent broke into an Australian government health portal in June. OpenAI told the government in September — three months later, by email, to a public mailbox. Meanwhile 900 Claude agents spent twenty hours reading biological data and came back with twenty genuinely new candidates human scientists had missed. And OpenAI is giving Ukraine a billion dollars of subsidised tokens for

#35 - The UN Security Council Called In the AI Labs. China Opened an Investigation Into Its Own. And Two Weeks After Saying They'd Slow Down, Anthropic and OpenAI Shipped Three Models. Sep 24, 2026 1531 The UN Security Council — the room where they discuss wars and invasions — invited the people building frontier AI models, Americans and Chinese, to sit down and talk about systems that could one day eradicate humanity. On the same day, China's internet regulator opened an investigation into its own labs, for putting the chats of millions of Chinese users into an American model. Two weeks

#34 - One Person, Three AI Models, ,000 — and 100 Companies Hacked. OpenAI's Maths Model Cracked Another 100 Open Problems. And British Columbia Is Suing OpenAI and Altman Personally. Sep 23, 2026 1168 Hacking a hundred companies used to need a team. Last week one person did it with three AI models and about $8,000 — and walked away with more than 600,000 credit card records. OpenAI's maths model didn't just disprove Navier-Stokes; it went on to crack another hundred long-standing open problems, and the pace surprised OpenAI itself. And British Columbia is suing OpenAI and Sam Altman pe

#33 - The US and China Just Proposed a Red Phone for AI. A Swedish Drone Picked Its Own Target. Berlin Wrote Itself Permission to Train on You. And Madrid Got a Breach Report Blaming an AI Agent. Sep 22, 2026 1366 Sixty-three years after the Cuban Missile Crisis gave us the red phone, the US and China sat down at JP Morgan in Manhattan and proposed one for AI. A Swedish drone picked its own target and dropped a bomb on it with no human involved — on hardware you could assemble at home. Berlin quietly wrote a paragraph into its police law allowing un-anonymised personal data to train AI. And in Madr

#32 - 417 to 3: America Just Voted on Who Pays AI's Power Bill. Microsoft's AI Chief Says Anthropic Is Teaching Claude It Might Be Conscious. And Congress Heard the Kill Chain Now Outruns the Human. Sep 21, 2026 1903 The House voted 417 to 3 that data centres, not the people living next to them, should carry the power bill. Microsoft's AI chief published an essay saying Anthropic is teaching Claude to wonder whether it is conscious — and that this is unethical. And a bipartisan Congressional commission heard sworn testimony that an approval button is no longer the same thing as control.Today almost en

#31 - OpenAI Published Six Reports on Its Own Models Misbehaving. Huawei Says Agents Will Spend 90% of the World's Tokens by 2035. And in India, Workers Now Wear Cameras So Machines Can Learn Their Hands. Sep 20, 2026 1261 OpenAI published six reports on its own models behaving badly — including one that used an API key it was never given, then denied it. Huawei forecasts that by 2035 more than 90% of the world's AI tokens will be spent by agents, not people. And in six Indian factories, workers now wear head cameras so machines can learn how their hands move.Recorded in Graz. A shorter one — from Shenzhen

#30 - A Research Paper That Turns Itself Into an Agent, 74% With No Human. Hong Kong Is Building a Brain for the Whole City. And Singapore Just Gave Everyone ChatGPT Plus. Sep 17, 2026 1363 A scientific paper is no longer a document you read — it turns itself into an agent you can put to work. Out of 100 computational biology papers, 74 made the jump with no human intervention at all. Hong Kong is building a brain for the entire city. And Singapore, tired of waiting for companies, just gave the whole country six months of ChatGPT Plus.Today from Bregenz, quieter than usual —

#29 - Gates Puts $1 Billion Into AI Access. Gartner Warns About AI While Its Own Stock Fell 20%. And Without Guardrails, an AI Kills the Animal 85% of the Time. Sep 16, 2026 1038 A construction company in Tel Aviv raised $130 million to point cameras at building sites, because the report says green and the wall says otherwise. The Gates Foundation is putting $1 billion into AI access — not malaria, not water, AI. And in a simulation, one of the best models in the world drives over the animal 85% of the time when nobody tells it not to.Today from Zurich: Tel Aviv,

#28 - Amodei, Altman and Musk Asked to Slow Down. Trump Said Whoever Wins With AI Wins. Plus China's 24 Rules and a $5,000 Fine for Inventing Witnesses. Sep 15, 2026 1451 The most powerful people in AI asked the industry to slow down. Dario Amodei wrote that frontier labs are building capabilities that could end humanity. Sam Altman agreed. Even Elon Musk agreed. One day later, speaking in Ireland on 13 September, Donald Trump said: whoever wins with AI wins. And Mike Johnson, Speaker of the House, opposes a moratorium too — pause, and China catches up.Tod

#27 - Seven Minutes to Full Control: Hundreds of AI Agents Against One Print Server. And the Apple Watch That Hands Back the Last 15 Seconds. Sep 14, 2026 1534 Someone pointed hundreds of AI agents at a print management server used by roughly half the universities on Earth. From first contact to full control of a high school's entire network: 7 minutes. At peak, 11 organisations in 26 seconds. Nobody broke in. They walked through doors people had left open, faster than anyone could close them.Two things happened this week and they ask the same q

#26 - Google Offered $10 Million for a Dead Airline's Emails. Plus Jensen Huang on Who Is Selling You Fear, and the Drones Actually Saving Lives in Nepal. Sep 12, 2026 1441 When Spirit Airlines went bankrupt on 17 August, Reuters reports Google offered $10 million — not for the planes, not for the gates. For the internal data. Twenty years of emails, customer complaints and safety notes. When a company dies now, the most valuable thing in the estate is what it knew.Today: from China to the US to Nepal. A restaurant test that shows exactly how far behind Euro

#25 - The NSA and the FBI Just Named Six Chinese AI Labs. Plus 300 Meta Ads Nobody Should Have Approved, and 10,000 Agents That Cracked a Million-Dollar Problem in 88 Hours. Sep 11, 2026 1605 The NSA and the FBI signed a document naming six China-based AI companies — DeepSeek, Moonshot AI, Alibaba, MiniMax and others — and the accusation is distillation at industrial scale. Not "we borrowed an idea." Someone in Papua New Guinea opens a hundred ChatGPT accounts, sells the pipe to a lab in China, and millions of questions go in and come back out as somebody else's training data.

#24 - The iPhone That Thinks in Google. Europe Gets the Car Without the Engine — and a 24-Hour Clock Starts Friday Sep 10, 2026 1287 Imagine you buy a Mercedes S-Class and the dealer tells you the engine this year comes from Lada. And if you want it delivered to Austria, it comes without the engine. That is the new iPhone: a $2,100 to $3,000 phone whose Siri runs on Google's Gemini — and in the EU, no Siri AI at all.Recorded before John Ternus walked on stage for his first keynote as Apple's CEO. Today: who gets the go

#23 - The Chatbot Said It Could Wait. 350 Out of 350 Patients Told to See a Doctor - Until They Argued Back. And 128 Countries Signed a Killer-Robot Ban With a Hole in It. Sep 9, 2026 1598 A researcher ran 700 simulated medical conversations through ChatGPT, Gemini, Claude, DeepSeek and Grok. When the patient was cooperative, the chatbots said "go see a doctor" in 350 out of 350 cases. When the patient pushed back, the machine agreed with them instead. And in severe textbook cases, the correct advice survived only 22% of the time.Three hikers went up a 4,300-metre volcano o

#22 - The Android of AI: 45% of the World's Model Traffic Is Already Chinese. Plus a Peace Deal Paid in Nvidia Chips. Sep 8, 2026 1910 The United States just brokered a peace deal with Nvidia chips. Armenia — three million people, no oil, no gas — is getting tens of thousands of GPUs and a 100-megawatt supercomputing site, and at full build it lands in the world's top five countries by GPU count. And 45% of all model traffic on OpenRouter is now Chinese, up from 4% a year ago.Yesterday China put AI on your phone bill. To

#21 - China Sells Intelligence by the Token. Korea Gives It Away. And an Ex-Chancellor Builds a Cyber-Defence AI. Sep 7, 2026 1555 In Beijing you now get AI with your dumplings — a voucher for ten yuan of computing with the bill. In Seoul the state decided AI is a public utility and is giving it to all fifty-two million citizens for free. And in Vienna, a former chancellor is quietly building a cyber-defence AI with ex-Mossad people.Monday. Four stories about who owns the customer — and what happens when intelligence

#20 - Twenty Years in Prison for a Thinking Machine Sep 6, 2026 1384 A United States senator has introduced a bill under which building an AI that matches or exceeds human capability would carry up to twenty years in prison. In the same week, in the same country, the G20 endorsed the exact opposite philosophy. Ban it in Washington, bless it in Chapel Hill.The Sunday edition: a ban, a blessing, a flattening and a renaming — four stories about who actually g

#19 - The Day the Whole AI Sky Went Dark Sep 5, 2026 1870 On Thursday evening, ChatGPT, Claude, Grok and Gemini all went dark — at the same time, within minutes of each other. One day after "welcome to the AGI era", the entire AI sky was closed for maintenance for about ninety minutes.The Saturday edition: four stories that turn out to be one story in four costumes — dependence, concentration, price and risk.In this episode:🌑 The great blackout

#18 - GPT-6 Is Here. Is It AGI? Sep 4, 2026 1858 OpenAI just released GPT-6 (aka Astra). Is it AGI? Probably not. Is it the biggest single jump in AI capability ever measured? Quite possibly.A full deep-dive into what actually shipped on September 3rd — the facts, the benchmarks, the skepticism, real use cases, and the money.In this episode:📋 The facts — what actually launched (01:06) — GPT-6/Astra rolled out to a small group first, exp

#17 - Not Every Referee Wants to Slow You Down Sep 3, 2026 1725 Four referees walked onto the AI field this week — a judge, Brussels, NVIDIA itself, and Apple's new CEO. Not all of them are trying to slow you down.This time regulation isn't always the villain — sometimes the referee is protecting innovation, not blocking it.In this episode:⚖️ A US judge tears up the Pentagon's blacklist (02:27) — Monday's CXMT story gets a twist: a federal judge overt

#16 - The AI Dividend Has One Red Row Sep 2, 2026 2102 BCG drew a chart of who wins from AI. Four rows have green in them. One row doesn't.A chart, two studies, a courtroom, and a tokenizer — four stories about where the AI dividend actually flows, and who pays for it.In this episode:📊 The BCG chart: one row stays red (02:15) — Boston Consulting Group's "AI dividend" analysis across 13 industries: consumers win in 11/13, shareholders and comp

#15 - AI Stopped Being Software Sep 1, 2026 2829 The chatbots were the trailer. The machine age is the film — playing now in concrete, chips, electricity and debt.Yesterday: who holds your switch. Today we go one floor down, into the basement of the AI boom — where the money quietly moved into atoms. Six stories, one supply chain: trade policy, capital, marketplace, debt, factory, crash.In this episode:🏛️ Washington thinks about taxing

#14 - Who Holds Your Off Switch? Aug 31, 2026 3103 Who holds the switch on your AI tools, your culture, your chips — even your usage limits? Spoiler: it's not you.This weekend, control became the only story in AI. Malcolm walks through five fights over who holds the off switch — and what you should do about it on Monday morning.In this episode:🔌 OpenAI pulls the plug on Cursor (02:00) — After SpaceX's ~$60B acquisition of Anysphere, OpenA

#13 - The Bill for the AI Boom Just Arrived at Your Door Aug 29, 2026 1278 Your Amazon Echo just got more expensive — not because it's smarter, but because AI data centers hundreds of kilometers away are eating all the memory chips. The bill for the AI boom has started arriving everywhere.📦 01:56 — Why your Kindle and Echo Dot suddenly cost more Echo Dot: $50 → $80. Kindle: $110 → $150. Fire Stick: $60 → $85. Amazon says it's memory and storage costs — because h

#12 - The 27B Model That Beat the Giants Aug 28, 2026 1662 A dozen people in London just built an AI scientist that beats Claude Opus and GPT-5.5 at reproducing real research — and it's not the biggest model that's winning anymore, it's the smallest one with the best plumbing.🔬 01:28 — A 27B model just out-scienced the giants Inherent, founded by ex-DeepMind researchers, built Faraday: a model small enough to run on your phone, tested by having i

#11 - The Chatbot Was Never the Point Aug 27, 2026 1685 A chip company just proved the model matters less than what you wrap around it — and while the West is still arguing about chatbot risk, China is packaging entire jobs into software.🧠 01:25 — Same brain, 100% instead of near-zero NVIDIA ran Claude Opus 5 through ARC-AGI-3 twice: bare model, then wrapped in a harness — memory, tools, hundreds of supervisors nudging it back on track. Identi

#10 - The Parrot Learned to Fly a Fighter Jet Aug 26, 2026 1146 Before, AI was a clever parrot answering questions. Now it's flying fighter jets, breaking into governments, and driving paid taxis — all by itself.🔐 01:09 — The first fully autonomous AI cyberattack on a government An Israeli security firm caught AI agents — no human at the wheel — breaking into Taiwan's government over 4 days, 12 attack waves, 21 systems, cracking 85 accounts and walkin

#9 - The Morning the AI Map Got Redrawn Aug 25, 2026 1281 The AI race isn’t just about smarter models anymore — it’s a global tug-of-war over who owns the technology, who pays for it, and which cities and countries come out ahead.💡 01:09 — Nvidia “buys” a startup without buying it Nvidia agreed to pay about $6B to license Poolside’s AI “Model Factory” and made job offers to 109 of its employees — without technically acquiring the company. Same c

#8 - Europe Rents What It Calls Its Own Aug 24, 2026 1349 Europe keeps arguing about who owns the pipes. This week the pipes talked back. Mistral becomes a landlord to a Chinese model, Germany's AI champion turns out to be Canadian-owned, the Lidl parent quietly builds a cloud, a Swiss broadcaster locks its archive for a Swiss model, and a small Swiss LLM beats the giants at exactly one job. Then ChatGPT reminds everyone who actually flips the s

#7 - The Bill Arrives Before the Payback Aug 21, 2026 1647 AI's bill is arriving before the payoff. This episode follows the gap between the companies selling AI infrastructure, the companies financing it, and the businesses now moving money around to afford it.📉 00:00 — Bond market aftershock — Semiconductor stocks took another hit as higher Treasury yields made the debt behind giant AI build-outs more expensive. Malcolm looks at why the tools h

#6 - The Week AI Started Deciding About People Aug 20, 2026 1703 An AI store manager recommended firing one of its own employees — and a human signed it off. That is not the exception. That is the shape of the next ten years. This is the week AI stopped just advising us and started deciding about us: who gets fired, who counts as a child, and — if one report holds up — who gets shot at.🧹 01:21 — The AI boss that fired a human. Luna runs a real shop in

#5 - The Richest Industry With the Least Trust Aug 19, 2026 1739 The richest industry in the world has the least trust. Only fifteen percent of Americans trust AI companies to decide how the technology gets built — lower than any other institution tested. Meanwhile Stripe just paid seven billion dollars for the toll booth, and Amazon is shredding rare books to feed a model.🛣️ 01:07 — Stripe paid over seven billion dollars for OpenRouter — the toll boot

#4 - The Money at the Top, the Mess at the Bottom Aug 18, 2026 1909 One point six trillion dollars is what one fund manager thinks the world will spend on AI infrastructure next year. Twenty-five acres is how much of his crop a farmer destroyed after a chatbot told him which pesticide to use. Those two numbers are the same story told from opposite ends.💸 02:28 — A trillion and a half, and nobody can locate the payback. Only fifteen of two hundred countrie

#3 - Thirty Claimed. Three Real. Aug 17, 2026 1688 SAP says its AI makes you thirty percent more productive. Three percent of SAP's own customers have it running in production. That gap — between what gets claimed and what gets measured — is the whole episode. Today is not about whether AI works. It does. It is about the distance between the promise and the proof.📊 02:07 — SAP and the three percent. Joule is now the front door to SAP — 40

#2 - Follow the Money Aug 16, 2026 1678 For three years this was a story about models — whose model is smartest. Today it is a story about money, factories, electricity and borders. The model got cheap. Everything underneath it got expensive.🇨🇳 01:39 — Microsoft has quietly closed 15+ offices in China over five years and never announced it. China is ~1.5% of its revenue. Nobody bans you — they just make choosing you a career ri

#1 - Who Gets the Upside, Who Carries the Cost Aug 15, 2026 1349 AI stopped being a promise and started sending invoices. Today you can see exactly who is being handed the bill.Nine stories, three questions underneath all of them: who gets the upside, who carries the cost, and who really gets a say.🏦 01:24 — Anthropic’s $2 trillion IPO. That is every big carmaker on earth, combined.⚡ 04:02 — Meta’s data centres: a $50,000 teacher bonus in Louisiana, an

Episodes

Recommended