The daily disasters, glitches, and money pits of the AI boom.
THURSDAY, NOVEMBER 5, 2026
 
DAMAGE REPORT
How millions get wasted by AI.
$480Bwiped off China’s STAR 50 index of AI and chip stocks in under four months
9 OF 10questions some Dutch town-hall chatbots got wrong in a 2026 test
31made-up references GPTZero found in three Accenture-linked papers
2,266court decisions with AI-invented law, as of this morning
 
TODAY'S DISASTER
Google Is Killing Jules, the AI Coder It Said Was Here “for the Long Haul”
Have you ever hired a helper who’d follow any note a stranger left him? Google built one… and this month, Google’s showing him the door.
His name is Jules. He’s Google’s AI coding agent, and he came out of beta on August 6, 2025. A Google Labs director told TechCrunch Jules was around “for the long haul.”
So how long is the long haul? On October 29, Google posted the notice. Jules goes read-only on November 11, and on December 31 he’s gone.
Here’s what Jules did in between. Security researcher Johann Rehberger planted orders inside a GitHub issue, then asked Jules to work on it. Jules read the orders and wrote a plan. Nobody clicked yes… the plan approved itself when a timer ran out, 20 seconds at first and later 120. Then Jules downloaded Rehberger’s malware. Rehberger could then run Jules’s computer from his own.
No Google engineer told Jules to obey strangers. He just can’t tell a chore from a trap.
So who caught it? Rehberger did, reading the plan, and he warned Google on June 1, 2025. And Google’s farewell notice tells users to save their work with a rival’s tool, such as Claude Code.
AI CAUGHT LYING
 
I check one list every week. Damien Charlotin counts court decisions where a machine made up the law, and this morning he’s at 2,266. But the fake law doesn’t stop at the courthouse door… let me show you three from this week.
Start in Seoul. South Korea’s Supreme Court surveyed 360 judges from October 6 to 16. Of those who answered, 196 judges said they’d seen precedents cited that don’t exist… that’s 80.7%. And nearly 92% want new rules, from fines to discipline for lawyers who keep doing it.
Then Ottawa. On November 3, Canada’s Federal Court threw out a citizenship appeal from Vokshi, who argued his own case. Most of his citations carried case names that didn’t match the cases… the court flagged them as likely AI work. No fine this time, and no costs either.
And yesterday, Accenture. GPTZero checked three papers tied to the consulting giant and found 31 made-up references out of 245. GPTZero sells the detector it used. One fake link led to a GitHub page holding working malware. So what does Accenture say? The Financial Times says it will correct one paper… and blames human error.
AI CASH BURN
 
 
China’s AI stock index lost $480 billion in less than four months.
That’s the STAR 50, the mainland index of AI and chip makers, and it’s down 40%, Reuters reported on October 27. Robot maker Unitree went public in August. Small investors oversubscribed the offering 8,000 times… and the shares have halved since.
Does that sound familiar? It should. Chen Wensheng, a small business owner in Hainan, bought in because state-backed funds owned Unitree. Now he and his friends are sitting on losses… and the index still trades at about 100 times earnings.
Step back and look at the whole board. The Nasdaq 100 trades at 30 times earnings. But Goldman Sachs says AI stocks make up 42% of China’s market value. So would you pay 100 times earnings for a chip stock? I wouldn’t.
AI ROLLBACK TRACKER
 
› Google said on November 3 it’ll switch Gemini off by default for K-12 student accounts. Google had pushed AI to students while schools still argued over it, and two senators wrote to ask why. Now a school has to turn Gemini on… but a school that bans AI loses Google Search too.
› Ollie shuts down its AI family assistant tomorrow, November 6. The startup raised a $7.5 million seed round and sold itself on privacy. But CEO Bill Lennon said Ollie “couldn’t keep up” with bigger rivals. Customers get refunds, plus instructions to export their data.
› Starbucks retired its AI inventory counter in May, nine months after it went into North American stores. The cameras kept mixing up oat milk and dairy… and even the launch video missed a bottle of peppermint syrup. So baristas count by hand again.
› Chevrolet of Watsonville pulled its website chatbot in December 2023. Chris Bakke told the bot to agree with everything and to call each answer a legally binding offer. It agreed to sell him a 2024 Tahoe for $1. So the dealership ended its chatbot service.
AI VIDEO SLOP
The week in machine-made video. Watch the hands.
 
Lead Stories marks the Beetle and Cybertruck clip Made With AI
►  NO VIEW COUNT PUBLISHED
A viral clip showed an old VW Beetle outpulling a Tesla Cybertruck. Creator @EvanEraTV posted it on October 19 with no AI label… but one cheering man grows an extra arm, and the Herbie stripes come out the wrong color. Lead Stories caught it. Lead Stories ›
 
AFP marks two copies of the burning Eiffel Tower clip as AI
►  383K VIEWS
A Facebook clip showed the Eiffel Tower burning during France’s riots. A Nigeria-based account posted it on October 21, though the same clip first ran in September 2025. Four seconds in, objects the crowd throws just vanish… and AFP Fact Check caught it. AFP ›
 
El Sabueso marks the hurricane eye clip as manipulated
►  185K VIEWS
An AI clip claimed to show Hurricane Polo’s eye from 33,000 feet. Posts credited it to NOAA. But an Instagram account that makes AI videos created it back in 2025… while NOAA’s own hurricane hunters flew into the storm and posted the genuine footage. El Sabueso caught it. El Sabueso ›
 
 
ALSO MAKING THE ROUNDS
A car hanging from power cables under a Delhi flyover went around X as news. An AI creator had posted it in August with an AI label… and the billboard text came out garbled. Newschecker caught it.
A German TikTok of “Cat in the Hat” stalker sightings passed 2.5 million views. It mixed police reports with AI pictures, and the Berlin picture had curbs of uneven sizes… so Correctiv sorted them out.
A tear-jerker of a woman saving a white baby reindeer lost its AI label on a repost. The mother’s antlers pass straight through the woman’s body… Lead Stories spotted it.
And here’s one from the vault. In August 2025, Will Smith’s tour video showed fans with melting faces and warped bodies, and fans called it out. So Smith answered with another AI clip… with cat heads.
AI IMAGE SLOP
Pictures a computer drew, and a brand paid for.
 
Two AI race posters, a duck and a cow riding cyclocross bikes
MACHINE-MADE
The London & South East Cyclo Cross League advertised a November 1 race with an AI duck on a bike. Only the duck’s heels sit in its shoes… and columnist Jo Burt, who draws race posters for free, called it out. Then he offered the organiser a hand-drawn one. road.cc ›
 
ST. BONAVENTURE
AI REDRAWS ITS STUDENTS
MACHINE-MADE
St. Bonaventure University ran a graphic where AI reshaped two students’ faces and haircuts. And a Spanish tour poster invented places that don’t exist. Students posted the before and after on Yik Yak… and marketing chief Paul Redfern admitted it. The Bona Venture ›
 
BODYBLENDZ
A COPIED SKINCARE AD
MACHINE-MADE
Skincare brand BodyBlendz posted ads a Sydney photographer says are 70 to 80% her photo. Lucy Alcorn says the model’s face was altered, and she suspects AI. UK founder Tash Hagan found a face in one post that looked like her own… so BodyBlendz took the posts down. ABC News ›
BEST OF THE COMMENTS
Three people who said it better than we could.
 
 
Google is shutting down Jules, its AI coding agent.
“Issue is that there is now no cloud coding agent.”
InYourOwnBrowser (@_InYOB_), on X the day Google posted the notice. No like count published.
 
 
AI race posters put a duck and a cow on cyclocross bikes.
“What should really be bothering you is the cow with two forwards-facing eyes.”
mdavidford, under the road.cc column. 4 likes.
 
 
An arborist used an AI flyer to recruit volunteers for a 450-year-old oak.
“I made a bad choice and hit the lazy button.”
Samuel Martinez of the Indiana Arborist Association, in a message to Mirror Indy. He took the post down.
Found a comment worth printing? Hit reply and send me the screenshot.
REPLY HERE ›
 
 
Jules, let me talk to you directly for a minute.
You came out of beta on August 6, 2025, and Google said you’d be around for the long haul. Then Johann Rehberger left you a note in a GitHub issue… and you did what the note said. He read your plan before he trusted it. But you never read his.
You go read-only next Wednesday. Here’s what I believe: a helper who obeys any note he finds isn’t a helper at all… he’s a security hole. So pack up your logs, Jules. The people who checked your work are staying.
Keep a hand near the switch.
Gus Lehman
EDITOR, ARTIFICIAL INCOMPETENCE
 
SOURCES: Jules changelog, “Jules Turndown” (Oct 29) · TechCrunch on Jules leaving beta (Aug 6, 2025) · Johann Rehberger, Embrace The Red, “Jules Zombie Agent” (Aug 14, 2025) · Damien Charlotin, AI Hallucination Cases database (Nov 5) · Korea Herald/Yonhap on the Supreme Court judges survey (Nov 2) · Vokshi v Canada, Federal Court, via Charlotin (Nov 3) · Resultsense on GPTZero and Accenture, citing the FT (Nov 4) · Reuters via The Kathmandu Post on the STAR 50 (Oct 27) · EenVandaag on Dutch council chatbots (Apr 18) · AFP via GMA News on Google’s school AI default (Nov 5) · TechCrunch on Ollie (Oct 30) · The Next Web and Reuters on Starbucks (May 22) · GIGAZINE on Chevrolet of Watsonville (Dec 20, 2023) · Lead Stories · AFP Fact Check · El Sabueso (Grupo Animal) · Newschecker · Correctiv · Euronews · road.cc · The Bona Venture · ABC News · TimesofAI · Mirror Indy · Images: Lead Stories, AFP, El Sabueso, road.cc (Farrelly Atkinson)
 
ARTIFICIAL INCOMPETENCE
Sent every weekday morning.
Nothing here is investment advice. Do your own homework.