Twenty-six new bacterial species turned up in the cleanrooms where NASA assembled the Phoenix Mars Lander, rooms sterilized by reduced humidity, chemical scrubbing, and radiation, built to leave nothing alive that could hitch a ride to Mars. What survived carries genes for DNA repair under radiation, spore shells, sticky biofilms that shrug off cleaning agents: organisms defined by their fit to the sterilization protocol, not by anything NASA cared about. Several of them now look industrially useful on Earth — one strain makes an antimicrobial polymer used in food preservation, another makes zeaxanthin, an antioxidant good for eye health, another synthesizes anticancer compounds. The barrier meant to guarantee absence became a bioprospecting filter for a value it was never built to notice.
James O’Claire worries that Anthropic and OpenAI are running the same sterilization protocol on the model market: price frontier access like a luxury good, maybe lean on China-fears to get open weights banned outright. If it works the way NASA’s cleanrooms worked, it won’t produce absence. It’ll produce DeepSeek V4 and Xiaomi’s Mimo, models whose entire fitness is adaptation to that pricing barrier, and the byproduct won’t be a cheaper frontier model — it’ll be something useful for a reason the barrier never priced in: auditable, self-hostable, unowned by the thing trying to exclude it.