OpenAI is allowing four independent auditing organizations, including Apollo Research and METR, to access its unreleased AI models, source code, and training data. This initiative marks the first time OpenAI has granted such direct, on-site access to its systems. The audits aim to identify potential safety issues and misbehavior, though OpenAI retains the ability to redact or delay the publication of audit findings, raising questions about the true independence of the oversight. AI
IMPACT This move could set a precedent for AI safety auditing and increase transparency in model development.
RANK_REASON Significant policy change by a major AI lab regarding safety and oversight. [lever_c_demoted from significant: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →