How LLMs Memorize Phone Numbers (and How Labs Stop It)

Why language models can spit out PII verbatim, the research behind it, and a tiny canary-string eval you can run on your own fine-tunes.