SAIL-ON: studying the "science" of AI that adapts to the unexpected — a federal contract to RTX BBN Technologies (USAspending)

SAIL-ON (Science of Artificial Intelligence and Learning for Open-World Novelty) is a program to research and develop the underlying scientific principles, general engineering techniques, and algorithms needed to create AI systems. "Open-world novelty" means situations or rule changes an AI never saw during training. Many AI systems perform well within the distribution of data they learned from, but become unreliable the moment they step outside it — and confronting that weakness head-on is what SAIL-ON is about.

Why it matters: real-world AI inevitably runs into moments where its training assumptions break — a changed environment, an unexpected input, a new behavior from its counterpart. Whether an AI can detect such novelty itself and keep performing is a foundational question that bears on reliability everywhere from self-driving to defense systems. The point of this kind of basic-research contract is not to buy a finished product but to use public funds to grow the "science" of AI that everything else rests on. The specific technical content or results of this award are not in the source description, so we do not speculate here.

Viewed more broadly, this is one example of U.S. research investment at the headwaters of advanced AI. Basic research like SAIL-ON tends to become shared groundwork for many later applications — robustness, out-of-distribution detection, continual learning — rather than a single product. The recipient is a technology firm with a long track record in applied research, and the very pattern of such companies and institutions executing national research programs reflects a defining feature of the U.S. research ecosystem.