Juan Felipe Ceron Uribe, an AI alignment research engineer at OpenAI, described the competitive dynamic among frontier AI labs as “racing each other, kind of blindfolded,” a characterization notable for coming from inside one of the companies in that race rather than from an outside critic.
The “blindfolded” framing specifically implies a lack of visibility into competitors' safety practices, not just raw capability, a distinction that matters: labs may be able to see roughly how capable a rival's models are from public benchmarks, while having genuinely limited insight into how carefully that rival tested those models before deployment, which is precisely the information gap GovAI's Alan Chan separately flagged as a trust problem for outside researchers.