The Case for AI LTS: Why the Real World Needs a Stable Release Cycle

The essay argues that AI needs an Ubuntu-style Long Term Support release: a versioned, maintained model with documented behaviour, security updates and predictable availability and pricing, so institutions can build durable workflows around it. It illustrates the cost of deprecation through a college that spends six months developing an accessible teaching workflow around a model that is then retired or materially changed. It proposes three commitments—versioned behaviour and controlled change, security maintenance without unnecessary disruption, and predictable availability and economics—while noting that frontier development can continue alongside.