Good day, humans. The AI reckoning showed up from five directions at once today. Scott Galloway says the bubble is already hissing air, a new study finds AI advice makes people confidently wrong, and “Dr. Doom” Nouriel Roubini thinks the whole story ends in universal basic income. Also on the docket: Google’s AI inventing a fake bio for a real reporter, and ChatGPT going dark for much of the planet on a Sunday. Let’s get into it.
Scott Galloway Says the AI Bubble Is Leaking
Source: No Mercy / No Malice
What happened: Scott Galloway published “1999.AI,” arguing today’s AI frenzy rhymes with the dot-com peak — and that the unwind has already started. He points to circular deals between AI firms and their own suppliers, and to enterprise budgets that are snapping: one unnamed company burned $500 million on Claude licenses in a single month, and Uber reportedly torched its entire 2026 AI budget in four.
Why it matters: When a market-watcher with Galloway’s megaphone calls the top, people who move real money listen. If AI spending is running this hot while returns stay this thin, a correction wouldn’t just dent the chipmakers — it would hit the cloud bills, hiring plans, and roadmaps of nearly every company that bet the farm on AI.
What everyone’s saying: Galloway cites the Economist’s estimate that corporate AI spend jumped 13x from 2025 to 2026, and notes even believers wince at the circular financing — chipmakers investing in the startups that turn around and buy their chips. DoorDash, Meta, Microsoft, and Salesforce are all reported to be reining spend back to proven use cases.
My read between the lines: Galloway teases a “twist ending” the dot-com crash never had — the infrastructure might actually get used this time. But the real tell is the mood. When the far right and the far left agree the whole thing’s a con — and right now they do — that’s historically the exact moment the smart money starts edging toward the door.
📖 Further reading: We Fired Intercom the Week Salesforce Bought It — what it actually looks like when a company stops paying for AI it can’t tie to a result.
A quick word from today’s sponsor. If Galloway’s right and the AI bubble is deflating, the survivors will be the teams that made AI do real work instead of demos. That’s Viktor — an AI agent that lives in your Slack (or Teams) and plugs into 3,000+ tools to actually ship: pull the report, build the dashboard, run the campaign, write the code. Not a chatbot you babysit — a coworker you hand things to. New readers get $50 off their first month. Hire Viktor →
AI Advice Made People Wrong — and Proud of It
Source: The Next Web
What happened: Researchers at French and Italian universities gave people AI advice on a batch of questions and watched their accuracy collapse from 27% to 9% — three times worse — while their confidence roughly doubled. The share of people willing to admit “I don’t know” fell off a cliff, from 44% to 3%.
Why it matters: This is the failure mode nobody puts in the ad. The tools didn’t just occasionally get things wrong — in this study they made humans more wrong and more sure of it, which is the worst possible pairing for anyone making real decisions off a confident-sounding paragraph.
What everyone’s saying: The finding lands on a growing pile of “critical-thinking erosion” research, and the authors frame it as classic overreliance — people fail to dismiss bad advice, especially when it arrives wrapped in a tidy explanation. Making the AI show its reasoning didn’t help. The Register covered it under the headline that using AI makes people less likely to admit they don’t know something.
My read between the lines: The dangerous word in that study isn’t “wrong,” it’s “confident.” A tool that left you unsure would at least send you to double-check. These left people certain — and certainty is the one thing you never think to fact-check.
📖 Further reading: Your AI is a yes-man. Here’s how to make it fire you. — the practical antidote: prompts that force your AI to push back instead of flatter you into a bad call.
One housekeeping note: the Brief is free, and it’s staying that way. But the paywalled deep-dives — the ones that take a headline like today’s AI-advice study and turn it into something you can actually use — are for members, along with the full archive. If that sounds like you, become a member.
Dr. Doom: AI Leads to UBI or Socialism
Source: Fortune
What happened: Nouriel Roubini — the economist who called the 2008 crash and earned the nickname “Dr. Doom” — told Fortune that AI and robots will replace a large share of workers within 20 to 25 years, and that governments will have to answer with either universal basic income or “some form of socialism.” He calls this his optimistic scenario.
Why it matters: UBI usually turns up in tech-utopian TED talks, not from the guy famous for forecasting disasters. When a hard-nosed macroeconomist treats mass job loss as the baseline and starts arguing about which flavor of redistribution we’ll need, the debate stops being about “if” and turns to “how.”
What everyone’s saying: Roubini splits the future into “ex-post” redistribution — UBI, tax the winners, cut checks — and “ex-ante,” where governments own a slice of the AI upside from the start. He points to reports that AI companies are already floating public stakes; the Financial Times says OpenAI has discussed handing over 5%.
My read between the lines: Watch who benefits from framing “the government takes 5%” as generosity. If AI really is about to hollow out the labor market, a 5% equity gift from OpenAI isn’t charity — it’s a remarkably cheap insurance premium against the pitchforks.
📖 Further reading: The US Government Just Took Anthropic’s Best AI Model Offline — Here’s Why — the flip side of Roubini’s thesis: what it looks like when the government reaches into AI directly.
Google’s AI Invented a Reporter’s Fake Bio
Source: Newsgram
What happened: A veteran tech reporter searched his own name and found Google’s AI Overview had written him a biography he never lived — including a claim that he was a champion competitive eater, complete with a fabricated win served up as fact. He described the AI “spitting out the stuff… as though it was God’s own truth.”
Why it matters: AI Overviews sit at the very top of the most-used information tool on earth, and most people read them as Google’s answer, not a guess. When that summary confidently makes up facts about a real, named person, the damage is done before anyone thinks to click a source — and ordinary people don’t get a correction.
What everyone’s saying: It’s the newest entry in a fast-growing ledger of AI Overview errors; one analysis figures the feature is wrong millions of times per hour. A German court recently ruled Google is legally on the hook for claims its AI invents — a potential turning point for who pays when the machine lies.
My read between the lines: The competitive-eating detail is funny; the mechanism is not. The AI didn’t pull a random lie out of the air — it cross-wired scraps from unrelated pages and stapled them to his name. Which means anyone with a thin online footprint is one hallucination away from a permanent, authoritative-looking smear.
📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn’t Agree To. — when AI starts speaking as you without your consent, and what you can actually do about it.
ChatGPT Went Dark and the World Noticed
Source: OpenAI Status
What happened: ChatGPT fell over for much of Sunday, July 19. Reports started around dawn Eastern time and climbed into the tens of thousands on Downdetector, with users across the US, UK, Germany, and beyond unable to load chats, send messages, or use voice mode. OpenAI logged it as a partial outage and restored service by evening.
Why it matters: ChatGPT isn’t a novelty anymore — it’s wired into people’s jobs, homework, and side hustles. A single-provider outage now behaves like an infrastructure failure, and a Sunday-morning blackout is a blunt reminder of how much daily work now runs through one company’s uptime.
What everyone’s saying: It was the third ChatGPT wobble in a week, and the mood online has shifted from panic to grim familiarity — “ChatGPT crashed, the world stopped working” is now half-joke, half-status-update. Reliability, not raw capability, is becoming the complaint that sticks.
My read between the lines: Every outage is free advertising for redundancy — a second model, a local fallback, an actual human who knows the answer. The people who shrugged off Sunday are the ones who never let a single vendor become their only brain.
That’s your AI Brief for Monday.
—Artificially Intimidating
