Elon Musk Wants AI Companies to Grade Each Other's Homework

Elon Musk wants AI companies to check each other's work. Speaking at the All-In Summit in Los Angeles on Monday, the xAI founder proposed that OpenAI, Anthropic, Google, Meta, and several leading Chinese AI companies test one another's models before releasing them to the public. The idea relies on a shared "test harness" — a common set of safety checks any rival could run. Musk's reasoning was blunt: "Instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns."

The proposal lands in the middle of a fight over speed. Anthropic CEO Dario Amodei has called for slowing AI development down, and researchers like Jacob Coxon and Evan Hubinger have publicly warned about serious risks. The White House has pushed back hard: President Trump called AI fears a "hoax," and China's Foreign Ministry dismissed slowdown calls as "fear mongering." Musk's plan tries to thread the needle — adding a safety checkpoint without asking anyone to stop building.

The catch is trust and secrets. Letting rivals examine your model before launch means handing them early access to valuable technology. No company has signed on yet. Firms would also have to agree on which tests matter, how results get shared, and when a discovered problem is bad enough to delay a release. Musk suggested logging test activity to catch intellectual property theft, but that doesn't solve the bigger question: would competitors actually cooperate in good faith?

For you, the payoff is quiet but real. More outside testing before launch could mean fewer embarrassing AI failures — wrong answers, privacy leaks, or biased decisions — reaching your workplace, your bank, or your kid's school. It's not a guarantee of safety, and it's not a law, just a suggestion from one executive. But an extra set of eyes before release is a meaningful idea in an industry that mostly grades itself.