rationale:

Strict Attribution Criteria To qualify, a general-purpose AI system must originate both the central idea and the main result. Specialist systems like AlphaFold do not qualify on their own Science's 2021 Breakthrough of the Year: AI brings protein ..., nor does AI assistance in analysis or optimization where a human drives the strategy https://www.bighatbio.com/bighat-anthropic-collaboration-report-10-27-2026. However, humans conducting physical bench work or writing up the final paper is explicitly permitted under this definition, and an AI authorship credit is not required Artificial Intelligence (AI) | Nature Portfolio.

The AI Capability Trajectory Current progress suggests general-purpose agents will likely originate almost all frontier theory by 2030. In late 2026, AI has already achieved headline-level results across mathematics, theoretical computer science, and physics, including a Navier–Stokes solution, the quasi-Riemann zero-free region AI edges closer to solving the Riemann hypothesis, one of ..., a refutation of the Exponential Time Hypothesis, and a claimed 4D Yang–Mills construction https://www.science.org/content/article/openai-just-changed-math-forever-exactly-how-no-one-yet-knows. With agent capabilities advancing rapidly and compute costs falling sharply, general systems will increasingly act as the primary ideators not just in theoretical fields, but across computational biology and chemistry AI edges closer to solving the Riemann hypothesis, one of ... Claude discovers a novel enzyme system.

Editorial Limitations and "Lumping" Science’s selections are deliberately field-balanced and heavily empirical. Historically, the Breakthrough list features observatories, archaeological discoveries, agricultural advances, and engineering milestones https://www.science.org/content/article/breakthrough-2024 Science's 2025 Breakthrough of the Year. Several slots in 2030 will inevitably go to legacy human-originated infrastructure or long-running clinical trials Science's 2023 Breakthrough of the Year: Weight loss ... Science's 2024 Breakthrough of the Year. Furthermore, Science often consolidates multiple AI scientific achievements into a single thematic item, as it did in 2025 Of all the breakthroughs in 2025, what topped Science's list? Science's 2025 Breakthrough of the Year. This editorial grouping behavior acts as a structural ceiling, meaning multiple paradigm-shifting AI discoveries might only secure one or two official list entries.

Validation and Experimental Bottlenecks While the theoretical output is staggering, it is not without friction. Many major 2026 claims remain unrefereed, and some face priority disputes How an AI math breakthrough ignited a controversy or verification issues in their formalizations Navier-Stokes lost in translation: Why Lean verification of .... In the physical sciences, wet-lab rounds still take weeks, and clinical trials take years. Although AI is accelerating these loops, human-efficacy breakthroughs and major experimental discoveries in 2030 may still largely rely on mechanisms or leads that originated from human researchers years prior. Considered alongside how AI integrates into different scientific fields, this contrast between rapid theoretical progress and slower physical-lab validation favors a smooth spread of outcomes rather than a sharp divide between limited adoption and near-total AI origination.

Summary of Expectations A median of 4.1 items, or approximately 4 qualifying entries, reflects a realistic balance: 1–3 theory or computational results, 1–2 AI-directed lab discoveries, and the remainder allocated to human-led empirical science. The central half of the distribution runs from 2.5 to 5.7 items, reflecting the different rates at which AI-originated ideas can become validated breakthroughs across fields. The lower end of the distribution, with a 10th percentile of 1.5 items, accounts for scenarios where Science aggregates all AI advances into a single "AI scientist" entry or if major AI claims fail independent verification Working with academic communities to build scientific ... https://www.anthropic.com/research/early-progress-drug-development. The long right tail, reaching 7.7 items at the 90th percentile, captures a rapid-transition scenario where automated labs become so pervasive that AI agents originate the core ideas behind the vast majority of experimentally validated breakthroughs, leaving Science little choice but to feature them across multiple distinct fields. These fractional quantile positions describe uncertainty over a count whose realized value will be a whole number.