OpenAI Publishes Research Acceleration Metrics on AI Agent Uplift
OpenAI has released quantitative data on how coding agents are accelerating its internal AI research cycle, showing autonomous task completion rising from 23% to 46% for two-to-four-hour tasks between March and October 2026, alongside a 77% increase in documented Safety and Alignment research effort. For defenders and AI governance practitioners, this is the first substantive public dataset on recursive self-improvement (RSI) indicators from a frontier lab, providing a concrete baseline against which oversight frameworks can be calibrated. Residual gaps include the absence of standardised cross-industry RSI metrics, unresolved questions about human oversight adequacy as agent success rates continue to climb, and the maturity required to translate internal transparency into actionable external governance.