Good day . A Meta bot got stuck pressing “remove from cart” on one artist’s shop, a vibe-coded AO3 clone lasted a single day, and OpenAI and Anthropic ran election war rooms. Plus Mamdani on Sam Altman and how workers across Asia and Africa really use AI.
Meta Bot Hits Artist’s Shop 550,000 Times Business Insider
What happened: San Diego children’s author and illustrator Chelsea Paulus told Business Insider that a Meta bot hit her website more than 550,000 times and overwhelmed her hosting. Fast Company reports the bot, Meta-ExternalAgent, got stuck in a loop on her site’s “remove from cart” function, making more than a hundred hits a minute. After she blocked it, a second Meta bot, Meta-WebIndexer, pushed her CPU to 100% and crashed the site.
Why it matters: Paulus told Business Insider she edited her robots.txt file and put up a firewall, but “those can also take 24 hours to take effect, and I needed to stop the bots immediately.” The fix that finally worked was code from her host, pasted into a hidden file on her site. Paulus, who used to work in web design, called it “an incredibly stressful week, even with some of my technical knowledge.”
What everyone’s saying: Meta’s answer: “website managers can decide how our crawlers interact with their sites through tools like robots.txt,” a spokesperson told Business Insider, adding that the company respects “the preferences site owners set.” Fast Company ties the episode to the scrutiny Meta faces over announced changes to its terms that, in its words, “essentially, allow AI crawlers to run wild across business accounts.” Three days ago we covered the one robots.txt line four labs will honor.
My read between the lines: “Just use robots.txt” is true the way “just lock your door” is true: it helps, and it works after the thing has already walked in. The machine built to read the whole internet got trapped by a shopping cart button, and the cost landed on a woman with four websites on one hosting plan. On Monday we covered Amazon’s stance on Meta’s Muse. Small shops don’t have Amazon’s bouncers.
📖 Further reading: AI Agents Can’t Find Your Business. I Tested Mine. — the other side of the bot problem: why the big sites are blocking agents, and what a bot actually sees when it visits a small business’s website.
Chelsea Paulus is a working artist who got dragged into server logs and hosting tickets when she should have been drawing. The same thing happens to every small business: the unglamorous stuff piles up while you’re busy doing the work you’re actually good at. Marblism sells AI employees that work inside the tools you already use, and Eva is the executive assistant: your inbox, your calendar and your meeting notes, checking in with you when she needs an approval. No prompting skills needed. New users get 10% off for life, and there’s a 7-day money-back guarantee. Meet Eva →
Vibe-Coded AO3 Clone Dies in One Day 404 Media
What happened: ao3.life, a vibe-coded site that scraped fanfiction from Archive of Our Own, reposted it and charged for AI-generated audiobook versions, went down Monday evening. Its creator said it received more than 2,000 takedown requests in the single day it was up, and that “a swarm of AI Agents controlling 40 virtual machines” did the scraping, 404 Media’s Samantha Cole reports.
Why it matters: The clone appeared after AO3 itself went down Thursday following a security incident. The Organization for Transformative Works, which runs AO3, says it could be down for as long as four weeks, with no data lost. AO3 hosts nearly 18.5 million works. Authors who had “archive locked” their stories to keep out bots reported the lock didn’t stop it: “I had everything archive locked on ao3 and it’s still all here,” one wrote on Bluesky.
What everyone’s saying: Readers and authors responded with widespread backlash on social media. The creator, who signed the apology “a 17 year old from India that thought they were doing something good,” said the site drew 126,592 unique users in 24 hours and sold “like 11 audiobooks.” Cole also notes some readers are now trading Google Drive links to works downloaded earlier, which alarms authors too.
My read between the lines: The takedown process asked authors to prove ownership by leaving a note on their AO3 profile. AO3 was down. That is an opt-out system whose proof step ran through the site the scraper had just replaced. The 91% author payout was a promise to hold money “indefinitely” for people who never agreed to be paid by a stranger. It’s a startup pitch that skipped a step: asking.
📖 Further reading: The Font That Beat AI for About a Week — another anti-AI shield that failed in public, and what it says about the doors we leave open.
The daily Brief is free and always will be. Members get the paywalled deep-dives behind the headlines, like what happened when seven AI bots ran a business for 72 hours, plus the full archive. Become a member →
OpenAI and Anthropic Run First Election War Rooms The Hill
What happened: For the first time, OpenAI and Anthropic ran election “war rooms” for Tuesday’s midterms. The New York Times reports a team of more than a dozen people worked around the clock in a conference room off the cafeteria at an OpenAI office in San Francisco, while Anthropic put analysts in cells in San Francisco, New York and Washington. Semafor first reported Anthropic’s.
Why it matters: OpenAI’s head of civic and strategic risk, Josh Lawson, told the Times that “The most available technology at the time in 2024 was text,” and “we’re not just monitoring text anymore.” The threats now include deepfakes, cyberattacks and influence operations that can be launched “with a series of prompts.” Yesterday’s lead was chatbots giving voters wrong dates; this is the staffed response.
What everyone’s saying: The Hill reported on Tuesday afternoon that ChatGPT had pulled more than 16.5 million times from Democracy Works, the nonprofit supplying voting logistics, per its second story. Semafor notes the war-room concept dates to 2018, when Meta stood up the first one, and that Google’s AI Mode failed last week to say who won the 2020 election.
My read between the lines: A dozen people in a room beside the cafeteria, guarding a product the Times says has about a billion users. The technology scaled; the safeguard is still humans watching a screen. Former Meta public policy director Katie Harbath told Semafor war rooms exist for “the things we can’t necessarily predict going wrong.” That is the honest job description: nobody can name the failure ahead of time.
Mamdani to Wired: No, I Don’t Trust Altman WIRED
What happened: In a WIRED interview published Election Day, New York City Mayor Zohran Mamdani said he worries AI is a bubble that will pop, and, asked whether he trusts “the Sam Altmans of the world,” answered “No.” He said there are “far too many similarities between the housing bubble and the state of AI in our economy today.”
Why it matters: Mamdani has supported New York State setting a one-year pause on data center construction. His complaint is opacity: these companies operate at “a level of opacity that renders it almost impossible for us to make an informed conclusion as to how much this query is actually costing us as a society.”
What everyone’s saying: He compared the industry’s pitch to Facebook in 2008, arguing “they should be able to regulate themselves.” On deepfakes, he said family members send him AI videos of himself believing they’re real, including praise for a 45-minute debate with a European member of parliament that never happened. “There is a real crisis of faith as to what is real right now.”
My read between the lines: A politician answering “do you trust Altman” with one syllable, then handing the question back to the interviewer, is notable by itself. What he’s really pushing against is the idea that AI is “so complicated” the only rational answer is to trust the builders.
Workers in Asia and Africa Quietly Rewire Jobs Rest of World
What happened: Rest of World profiled workers across Asia and Africa using AI to get daily work done. A Lagos makeup artist, Nureeyah Ogunmuyiwa, calls ChatGPT her “most reliable business partner” and used it to build her first hiring plan. A Bengaluru cafe owner, Chinmayee Mishra, built agents to calculate recipe costs and compare prices. A director in Zhengzhou, Wendi Ma, now makes minidramas entirely with AI video.
Why it matters: Ma says a series that once took about $200,000, a crew of 50 and two months now takes about 10 people, roughly $20,000 and under a month, with Seedance charging about 12 to 15 cents per second of video. The people generating the clips are nicknamed “lottery masters.” “It has already changed everyone at the bottom of the industry,” he told Rest of World.
What everyone’s saying: Rest of World’s framing is that the real AI economy is being built personally and locally, outside Silicon Valley pitch decks. Ogunmuyiwa hasn’t paid for any AI tool: “I have invested my time but not my money.” Mishra pays about 2,000 rupees ($26) a month for ChatGPT Plus and considers it one of her best business investments.
My read between the lines: Mishra’s rule beats most corporate AI policies: “If it’s a calculation based on my data, I trust it. If it’s subjective… that still has to be my intelligence.” She also turned down QR menus and customer profiling. Note who is automating what: the cafe owner automated the paperwork to keep the humans, the studio automated the humans to keep the pipeline.
📖 Further reading: Claude for Small Business Just Hired You 44 Employees. You Only Need Four of Them. — the owners in this story built their own workflows; this is Anthropic’s ready-made version, and which four skills are worth opening first.
That’s your AI Brief for Wednesday.
—Nicholas from Artificially Intimidating
