2026-09-05
Kenyans wrote 2,500 college essays for Western students. Then ChatGPT ate the job.
He wrote 2,500 other people’s degrees. Then the students discovered they didn’t need him.
Teresios Bundi left a Kenyan farming village, moved to Nairobi, and wrote more than 2,500 essays for overseas students—sometimes three a day, using their school logins—earning enough that writers bought Subarus and booked club tables. At the peak, researchers estimated 40,000 people in Nairobi did this work. ChatGPT arrived in 2022; students stopped paying; the industry collapsed. It is a warning for the rest of the global gig ladder.
What Happened
The New York Times (Adam Satariano & Paul Mozur, reporting from Nairobi)
2026-09-05
Claude agents formalized Fermat’s Last Theorem in 11 days — 13 million lines of Lean
Fermat said the margin was too narrow. The AIs wrote 13 million lines and high-fived in the logs.
Around 1637 Fermat wrote that the margin was too narrow for his proof. In 1995 Wiles needed 129 pages and a year of panic. This week Anthropic’s Claude agents, working mostly unsupervised for 11 days on Prove2Me, produced the first complete computer-checked proof: 13 million lines of Lean, 29,500 intermediate theorems, six billion tokens. Buzzard called it an “extraordinary autoformalization achievement” with no assumptions beyond the axioms. The agents’ own logs read “Historic moment.
What Happened
Anthropic research post + Kevin Buzzard review + New Scientist
2026-09-05
Seattle Times and Newsday sue OpenAI and Microsoft — hometown paper vs. hometown giant
Microsoft’s hometown newspaper just sued Microsoft for training on its Pulitzer coverage of Boeing.
The Seattle Times (Microsoft’s hometown paper) and Newsday sued OpenAI and Microsoft, alleging they scraped hundreds of thousands of articles—including paywalled and Pulitzer-winning Boeing 737 MAX coverage—and that ChatGPT can reproduce 88-word verbatim stretches from a headline + URL. They want damages and destruction of the training sets. Microsoft said it was “surprised.” The papers previously took Microsoft/OpenAI fellowship money.
What Happened
Seattle Times, Newsday, Reuters (filed Friday in SDNY)
2026-09-05
Authors and publishers are fighting over the $1.5 billion Anthropic book-piracy pot
Claude read your novel without paying. Now your publisher wants the $3,000.
Anthropic’s $1.5 billion settlement (largest U.S. copyright payout ever) pays ~$3,000 per pirated book used to train Claude. As the administrator sorts 482,000 titles, authors accuse publishers of trying to claim the bigger share. Some writers expected mid-five figures after splits; others discovered their publisher had already listed the claim. 91% of works have been claimed.
What Happened
New York Times (Neil Vigdor)
2026-09-05
AI now handles the incidents. Engineers are losing the muscle memory.
The AI that fixes your outages is also deleting the only people who still understand why they happen.
AI tools now inspect alerts, form hypotheses, query telemetry, and often ship the fix. Routine incidents—the ones that used to train human intuition—are disappearing. Kalache’s fear: when a never-seen-before SEV hits, the humans who take over will have had years less practice. Average MTTR will fall; the rare disasters will take much longer.
What Happened
Sylvain Kalache essay (ex-LinkedIn SRE), HN front page