Alongside them, a new generation of AI safety, alignment, and evaluation firms, including Lakera and Gray Swan, has emerged. Language solutions integrators are playing an increasingly important role, too. Slator's research shows that around 10% have established dedicated Data-for-AI divisions. (Notable examples include TransPerfect, RWS, Welocalize, and Acolad.) Against this backdrop, what do frontier labs actually look for when selecting data-for-AI partners?