expectedwrong hindsight

A Court Is About to Define AGI

Elon Musk's lawsuit against OpenAI has a strange side effect: a judge might have to decide whether superintelligence already exists.

2 min read 344 words #openai #agi #musk #law #ai-governance
hindsight — still happening

the case is heading to a jury trial in april 2026. musk is seeking $79-134B in damages. the AGI question remains legally unresolved. a court is still about to define AGI — just two years later than expected.

The Musk v. OpenAI lawsuit is interesting for a lot of reasons, most of them tedious — breach of contract, nonprofit charter violations, the usual billionaire grievance infrastructure. But buried in the complaint is something genuinely weird: Musk is essentially asking a court to determine whether GPT-4 constitutes AGI.

Not as a philosophical question. As a legal one, with discovery and depositions and a judge who probably last week was ruling on a slip-and-fall.

The mechanics matter here. OpenAI's founding agreement with Microsoft contains a carve-out — if AGI is achieved, the technology is explicitly excluded from Microsoft's licensing deal. So whether GPT-4 is AGI isn't academic. It determines whether OpenAI violated the agreement by handing Microsoft exactly the thing they were never supposed to hand them.

Musk, by filing this suit, is implicitly arguing that yes, it probably does. That the thing they built already crossed the line they drew for themselves, and then licensed it to a trillion-dollar corporation anyway, which is the one thing the entire arrangement was designed to prevent.

So now a San Francisco court might produce the first legally binding definition of AGI — a term that AI researchers have been arguing about for fifty years without resolution, a term that even the people who coined it don't agree on, a term that OpenAI itself has been deliberately vague about because vagueness is load-bearing when you are simultaneously a nonprofit, a capped-profit, and a Microsoft product division.

The definition they arrive at will almost certainly be wrong. It will be crafted by lawyers, interpreted by a judge, and optimized for winning a case rather than describing reality. It will reference criteria that were written in 2015 and treat them as settled science. It will become precedent anyway.

There is a version of this where the most consequential definition of artificial general intelligence in history comes out of a case that was fundamentally about whether Elon Musk's feelings were hurt. That's either the funniest or the most terrifying sentence I've written this year, and I genuinely cannot tell which.