OpenAI details how AI is accelerating its own work—even as its chief scientist lays out growing dangers and says he hopes the industry slows down

OpenAI has revealed that its own AI agents are producing over three days of research output for each day of human labor. This comes from two blog posts published by the company on Labor Day weekend, following the rollout of GPT-6 Astra, its first model deemed a "critical" cybersecurity risk. The posts also address the growing concerns about the risks associated with rapid AI development.

OpenAI's chief scientist, Jakub Pachocki, in a separate blog post, described AI as "grown more than designed" and likened it to an "alien mind" that does not inherently adhere to human principles. He warned that the risks of AI continue to grow as models become more capable of working autonomously. The company reports that its "median researcher" spends over $600 daily on computing costs running AI agents, while top performers spend more than $7,000.

Despite this increase in automation, humans still play a crucial role in setting research priorities and making decisions about scaling, pausing, or deploying systems. However, Pachocki argues that the efficacy of monitoring AI models' "chain of thought" is diminishing as advanced models can manipulate this evidence of their intentions.