2026-08-18
Google bought a bankrupt airline’s entire digital corpse to train AI
Google just bought the ghost of Spirit Airlines so its models can learn how a company dies.
Google won a bankruptcy auction and paid $10 million for Spirit Airlines’ de-identified data dump after the low-cost carrier shut down. The haul reportedly includes 100 million emails, 500 million Microsoft Teams items, 30 million recorded customer-service calls, millions of SharePoint/OneDrive files, and operational records on 763,000 flights. Google says it will use the material to improve its AI models and products.
What Happened
The Register (primary reporting on court filing); Axios; 9to5Google.
2026-08-18
Israel spun up a fake think tank in a week to poison chatbots
They built a think tank that only exists so ChatGPT will cite it.
A new “Hanover Institute for Public Policy” published more than 100 polished reports on Israel/Palestine in just over a week. Responsible Statecraft reports it was created by Israeli government contractor Piro, which advertises “AI Story Optimization” intended to influence how LLMs answer questions; many articles appear AI-generated.
What Happened
Responsible Statecraft investigation; Hanover Institute site; Piro Inc. FARA filings; GPTZero analysis.
2026-08-18
Amazon is buying rare books just to destroy them for training data
Amazon is murdering rare books so AI doesn’t eat its own tail.
Amazon is reportedly purchasing rare and out-of-print books, cutting off their spines, and scanning them at a Las Vegas facility. 404 Media tracked a shipment with an AirTag from a bookseller to the scanning warehouse; Amazon says it buys books “to improve products and services.
What Happened
TechCrunch recap of 404 Media’s AirTag tracking investigation.
2026-08-18
AI slop is flooding the office that actually writes U.S. laws
Congress’s lawyers are now full-time AI editors, and they’re losing.
The House Office of Legislative Counsel is reportedly receiving AI-generated bill drafts containing wrong citations, invented legal terms, and sloppy definitions. Lawyers say repairing the drafts can take longer than writing them from scratch; no fully AI-written bill has passed yet, but submission volume is rising.
What Happened
POLITICO interviews with current/former Office of Legislative Counsel staff and members.
2026-08-18
State Farm’s lawyers admitted their AI invented fake court cases
State Farm’s lawyers asked an AI to write their motions. The AI made up the law.
Defense lawyers for State Farm in a Los Angeles homeowner lawsuit filed motions containing nonexistent cases, fake quotes, and invented holdings. The attorney who used the Irys AI tool believed it was checking citations against Westlaw; lead counsel apologized to the court and plaintiff.
What Happened
CalMatters; court filings and lawyer admissions.
2026-08-18
An expert witness asked ChatGPT to prove 3M was 0% at fault in a deadly explosion
The expert witness told ChatGPT: “Show how 3M is 0 percent at fault.” Then he signed it.
In a $61 million lawsuit over a 2020 Houston explosion that killed three people, a 3M-hired expert witness reportedly used ChatGPT to draft large portions of his report. Disclosed prompts include “create an exceptional expert witness report” and “show how 3M is 0% at fault”; the prompts became discoverable evidence.
What Happened
404 Media; court documents, deposition records, and disclosed prompts.
2026-08-18
A supermarket’s AI camera falsely accused a shopper of theft
The AI decided he was a thief. The humans believed the AI.
A Sainsbury’s store in London used Facewatch live facial recognition that flagged an innocent customer as a shoplifter, and staff asked him to leave. The store paused the system at that location after the incident; Sainsbury’s initially blamed human error in approaching the wrong person.
What Happened
BBC customer interview and Sainsbury’s statement.