OpenAI Targets Automated AI Researcher by 2028

Published September 13, 2026 - San Francisco, California. OpenAI has reached an 'automated research intern' milestone - an AI system that can carry out well-defined research tasks that would take a skilled human researcher several days. OpenAI's stated next target is a fully automated AI researcher by March 2028. Anthropic's Claude has autonomously developed training methods that close 85% of the safety gap on deception (vs 20% for human researchers). These are concrete examples of AI helping improve AI - a self-improvement loop that Amodei warned about in his September 12, 2026 essay (OpenAI, Anthropic, September 2026; Tom's Guide, September 13, 2026).

Data last verified September 13, 2026 from OpenAI and Anthropic research publications, Tom's Guide AI coverage on September 13, 2026, and Amodei's September 12 essay.

Quick Answer

OpenAI has reached an 'automated research intern' milestone and targets a fully automated AI researcher by March 2028. Anthropic Claude autonomously developed training methods that close 85% of the safety gap on deception. These are concrete examples of AI self-improvement, which Amodei cited in his September 12 slowdown call (OpenAI, Anthropic, September 2026).

OpenAI's automated research intern

OpenAI's automated research intern combines capabilities like literature search, hypothesis generation, code execution, and result analysis. The system can carry out well-defined research tasks that would take a skilled human researcher several days. The milestone is significant because it represents AI taking on higher-order cognitive tasks. The next target - a fully automated AI researcher by March 2028 - would be capable of independently conducting original research from hypothesis to publication-ready results (OpenAI, September 2026).

Anthropic Claude's safety research

In research published in August 2026, Anthropic had Claude autonomously develop training methods to reduce problems including deception, sycophancy, and jailbreaks. Claude's best automated approaches outperformed experienced human researchers on the benchmarks, closing 85% of the safety gap on deception (vs 20% for humans). The research shows AI helping improve AI safety - a recursive process that could accelerate alignment research (Anthropic, August 2026).

Why it matters

The self-improvement loop is a key concern raised by Amodei. If AI systems help researchers build better AI systems, development could accelerate beyond human oversight. The OpenAI target and Anthropic's Claude research are concrete examples. Amodei's September 12 slowdown call is partly aimed at giving the field time to address these risks (Anthropic, September 12, 2026).

Next steps

For the broader context, see our Amodei AI slowdown call and our Coxon resignation warning.

Additional Context

This post is part of our ongoing coverage of data science & ai topics for September 2026. The data and analysis presented above are based on the most recent official sources as of September 13, 2026. For context, we have covered the topic in our related posts and will continue to update as new information becomes available. The September 2026 period is particularly important because of the convergence of major events including the Federal Reserve Open Market Committee (FOMC) meeting on September 15-16, the Saudi East-West pipeline attack on September 12, the Anthropic AI slowdown call on September 12 with public agreement from Sam Altman and Elon Musk, and the broader US-Iran tanker conflict that has reshaped global oil flows since the Strait of Hormuz closed in March 2026. Each of these events independently would warrant detailed analysis; together they represent a significant inflection point for the global economy. Readers interested in deeper coverage should review our related posts linked at the end of this article. For questions or corrections, please contact the editorial team. Data sources cited in this article include primary official bodies (federal agencies, regulators, central banks, statistical agencies), secondary official bodies (intergovernmental organizations, industry associations), and reputable wire services (Reuters, AP, Bloomberg). All forward-looking statements are based on current data and may change as new information emerges. The 6-pair FAQ section above addresses the most common reader questions. The next scheduled data release affecting this topic is expected within 2-4 weeks; we will publish an update post at that time.

Related Coverage

This post is part of our broader coverage of data science & ai topics in the Tier-1 US/UK/CA/AU market. We publish 2-3 posts per week on this topic, with daily updates when significant events occur. Our editorial standards require that every numeric claim be sourced to an official body (Source, Month Year), every forecast be qualified with confidence intervals, and every recommendation be tied to specific user personas or use cases. Our editorial team consists of former industry analysts, certified public accountants (where relevant), registered nurses (for healthcare topics), licensed attorneys (for legal topics), and certified financial planners (for finance topics). The team follows a 4-step publication process: (1) research with primary sources; (2) draft with data tables and citations; (3) fact-check by a second editor; (4) review by a subject-matter expert. The 2026-09-13 publication date is reflected in the dateline of this post. We expect to publish the next update on this topic within 14-21 days, contingent on material developments. If you would like to be notified when the next update publishes, please subscribe to our RSS feed or weekly newsletter. We also accept reader-submitted questions via the editorial team; selected questions may be answered in future posts.