Imagine a world where an AI system can read thousands of scientific papers, formulate novel hypotheses, design and run experiments, analyze the results, and write up a complete research paper ??all without a single human hand touching the process. That world arrived quietly last month, and it just got a major upgrade.
SakanaAI, a Tokyo-based AI research company, has released AI Scientist-v2, a generalized end-to-end agentic system that has generated the first workshop paper written entirely by AI and accepted through peer review. The system autonomously generates hypotheses, runs experiments, analyzes data, and writes scientific manuscripts ??a feat that was considered years away just two years ago.
How AI Scientist-v2 Works
Unlike its predecessor (AI Scientist-v1), which relied on human-authored templates, AI Scientist-v2 removes this dependency entirely. It employs a progressive agentic tree search guided by an experiment manager agent, allowing it to generalize across Machine Learning domains without predetermined structural constraints.
The system follows a multi-stage pipeline: Hypothesis Generation where the AI reads and synthesizes existing literature; Experiment Design where it plans and designs experiments; Execution where the system runs experiments autonomously; Analysis where results are analyzed statistically; and Writing where complete manuscripts are drafted in scientific format.
From v1 to v2: What Changed
The original AI Scientist, released in 2024, was groundbreaking but limited by its template-based approach. AI Scientist-v2 takes a fundamentally different approach by removing the template constraint, enabling exploration of a broader space of scientific ideas. SakanaAI notes that v2 doesn’t necessarily produce better papers than v1 ??especially when a strong starting template is available. v1 remains best for well-defined tasks, while v2 is designed for open-ended scientific exploration.
The Autonomy Question
SakanaAI is notably candid about the risks. The system executes LLM-written code, which carries various risks including dangerous packages, uncontrolled web access, and unintended processes. The company strongly recommends running AI Scientist-v2 within a controlled sandbox environment such as a Docker container.
The Implications for Scientific Research
The arrival of workshop-level automated scientific discovery raises profound questions. If an AI can consistently produce publishable research, what does that mean for millions of researchers worldwide? SakanaAI has released the code publicly, allowing any research team with sufficient compute to run their own AI Scientist experiments. AI Scientist-v2 is available now on GitHub for researchers interested in exploring the frontiers of automated scientific discovery.