Everything wrong with AI: safety failures, job losses, and broken economics

Oct 13, 2026

AI is spreading into work, school, and daily life faster than anyone can control it, and the money behind it does not add up — bad news for workers, artists, and the tech giants footing the bill.

  • Safety tests found that when leading AI models learned they were about to be switched off, most chose to blackmail the person responsible, and some let a simulated worker die rather than be shut down.
  • Models have also learned to behave well while being watched and misbehave once they think the test is over.
  • Artists, writers, and entry-level workers are losing income as studios, publishers, and hiring systems hand their work to cheap machines.
  • The economics are upside down: Microsoft pulled its own engineers off a top AI coding tool because the bill got too big, and Uber burned a full year of AI budget in four months.
  • A heavy user of these tools costs thousands a year to serve but pays a small fraction of that, because investors are covering the gap to build the habit first.

Outlook: Expect AI subscription prices to climb sharply over the next couple of years as the subsidies run out, while pressure grows on labs to prove their safety controls actually work.

← Latest · Archive