OpenAI alignment research engineer Juan Felipe Ceron Uribe described frontier labs as “racing each other, kind of blindfolded,” while GovAI's Alan Chan told a Washington briefing that models inside labs “haven't necessarily gone through a bunch of safety testing” and that “internal safeguards have not been deployed” in some cases, adding that outside researchers “can't trust them completely” to disclose what testing actually occurred.
These are specific, falsifiable claims from people positioned to know, not general concern about AI risk in the abstract. The appropriate response from frontier labs, publishing exactly what safety testing a model underwent before internal deployment, is a low-cost way to directly address the specific gap Chan and Ceron Uribe are describing, if the labs choose to take it.