The growing concern of closed AI

Yesterday Anthropic released its flagship model, Fable, alongside restrictions intended to prevent its use for LLM development. They’ve validated a growing concern amongst many of us - frontier AI companies believe they alone get to decide who can study, understand, and ultimately shape one of the most important technologies of our time.

Today it’s model builders. The very people most capable of independently evaluating frontier systems are being discouraged from studying them. We should be cautious whenever institutions ask to be trusted without being scrutinized. Perhaps the day will come when businesses are strong-armed into paying steeper prices. Perhaps certain forms of inquiry are deemed too risky, too competitive, or too undesirable that certain groups are denied access under the guise of safety.

We have a civilization based on science and technology. If we are in a position where we cannot understand the science and technology, verify it independently, access it without the decree of a select few, we risk concentrating extraordinary power over the future of our civilization in the hands of an unaccountable few.

This is not the first time powerful institutions have attempted to control access to transformative technology. European monarchies and religious authorities imposed licensing regimes on printing presses, arguing that dangerous ideas required oversight, preventing challenging beliefs from being published and shared. AT&T controlled much of the telecommunications ecosystem through its network, equipment, and patent portfolio; most famously gatekeeping the transistor and suppressing broader innovation.

An a more optimistic note, many have voiced concern over this stance. The public does care deeply about open science and we hope Anthropic will alter their stance in consideration of this.