OpenAI launches embedded external safety assessments, but ex-employee says Slack access is missing

j_asminewang · x · 2026-11-17

OpenAI announced embedded assessments with Apollo Research, METR, Transluce and Goodfire, giving outside experts deep access to internal models and systems to investigate alignment and test safeguards.

Former employee sjgadler is sharply critical after reading the details: the access excludes Slack channels — where, per his own time at OpenAI, most context, debate and safety concerns happen. He argues OpenAI effectively reneged on its "employee-like" access pledge.

He adds that Google, Meta and xAI have said almost nothing since signing the White House Accord in September, which committed to partnering with external groups.

Related event: OpenAI Launches Embedded Assessments with Four External Safety Labs(19 posts)→

Original post →

More from Safety

Safety channel →