What happened
Computer scientists developed a method using large language models to automatically add monitoring sensors into railway control software, then check if the software behaves as designed when it actually runs. In practice, this catches unexpected failures during operation — the kind that slip through testing but cause real problems — by comparing what the software actually does against what engineers intended it to do.
Why it matters
Railway safety systems are some of the most critical software in the world, and this shows a path to catching failures that traditional testing misses, which matters because 'unknown unknowns' — things engineers didn't think to test for — are a real source of accidents.