- OpenAI says “entirely automating everything is not the future we want,” softening last fall’s goal of fully autonomous AI research by March 2028.
- A new post from CEO Sam Altman and chief researcher Jakub Pachocki frames a “tandem” between humans and machines instead.
- The two call for an international body that could slow frontier AI development if needed.
- The shift is a notably more cautious tone than OpenAI struck in 2025.
What Happened
OpenAI is backing away from the goal of fully autonomous AI research, instead describing a “tandem” between humans and machines, according to The Decoder. A new blog post from CEO Sam Altman and chief researcher Jakub Pachocki states plainly that “entirely automating everything is not the future we want.”
Last fall, OpenAI had set an ambitious target: by March 2028, build a fully autonomous AI system capable of conducting research on its own.
Why It Matters
The reframing matters because OpenAI set the pace for the “automate the researcher” thesis the industry adopted. Pulling back signals that even its most aggressive proponent now sees limits — a contrast with the safety-first positioning rivals have used, including Anthropic’s framing around Claude Opus 4.8 as its “most honest model.”
Technical Details
The post softens the 2028 milestone: rather than a system that fully automates research, OpenAI now says it may by then have AI working alongside human researchers. Altman and Pachocki also call for an international body with the authority to slow frontier development if risks warrant — a governance proposal, not a product claim.
Who’s Affected
AI researchers and policymakers are the primary audience. The call for international oversight lands amid active legislative fights over AI governance, including the Great American AI Act’s proposed state-law preemption.
What’s Next
The post is a statement of intent, not a binding commitment, and OpenAI did not detail how an international oversight body would be constituted or enforced. Whether the company holds to the more cautious framing as competitive pressure mounts is the open question.
Related Reading
- Alibaba’s Qwen3.7-Plus Turns a Multimodal Model Into an Autonomous Agent
- xAI Reportedly Trained Its Coding Models on Claude Outputs for Months
- Meta’s First Paid AI Product, Hatch, Could Cost Up to $200 a Month
- GPT-5.6 Is Expected Within Days — Markets Give It 89% Odds for June
- Three Companies Worth $4 Trillion Are All IPO-ing This Year
