Previously worked on safety at OpenAI, co-founded Guidelight nonprofit developing safety standards for AI companies, published scorecard on frontier lab control practices.
no scored calls yet — needs a stated position or a categorical verdict, with a matured window vs SPY
xAI has the weakest AI control practices of any frontier lab, lacking both third-party auditing and basic containment infrastructure, posing governance risk as its models scale.
Steven Adler's nonprofit Guidelight published a control standard scorecard assessing frontier AI companies on preventative containment practices, finding no company above 3/5 and exposing industry-wide reliance on reactive rather than preventative safety systems.
OpenAI ties for best safety score (C+) and has best containment transparency due to Hugging Face incident response, but still relies on after-the-fact detection rather than preventative controls against model misbehavior.