About 40% of AI-using scientists expect AI to complete full research projects by 2030

About 40% of researchers who actively use AI expect models to carry out full research projects entirely on their own by 2030, according to a report published by Anthropic1.

The finding, from a survey of AI-using scientists reported in the same study, offers a quantitative signal for how quickly the researchers closest to the technology believe autonomous discovery could arrive. The survey is part of a vendor-published study by Anthropic researchers that also analyzed 207,049 Claude and Claude Code research sessions across 34 scientific fields, recorded between August 2 and September 30, 2026, on consumer plans Anthropic1. The report captures survey expectations, not demonstrated capability, and Anthropic provides no independent test of autonomous research.

How scientists use the tools today

The session data reveals a mismatch between what scientists value most and how they actually use Claude. Data analysis dominates, accounting for about one-third of Claude sessions and nearly half of Claude Code sessions Anthropic1. Idea generation, which the surveyed researchers rank as the most important task for producing major findings, makes up a far smaller share of usage. On Claude Science, idea generation represents about twice the share of use that it does on Claude, suggesting specialist tools could shift behavior Anthropic1.

Usage also concentrates in fields where AI’s theoretical capability is highest. Computer science, mathematics and statistics show heavier relative use. Lab-based physical sciences and medicine record less, a pattern Anthropic attributes partly to physical bottlenecks and patient-facing work Anthropic1.

A hypothetical acceleration

Anthropic modeled a scenario in which AI automates all non-physical research work. In fields such as computer science and statistics, research output could rise as much as 15-fold. In bench sciences such as chemistry and biology, where wet-lab experiments remain a constraint, the modeled increase was three- or fourfold Anthropic1. The report does not attach a timeline to these acceleration estimates.

The report's researchers also conducted an accompanying survey1 of researchers who actively use AI. About 40% of those surveyed expect AI models to carry out full research projects on their own by 2030 Anthropic1. The survey focused on 528 researchers, mainly in life and health sciences, who already use AI, according to a summary on X by @heynavtoor2. The X post also states that 38% of those surveyed believe AI can already match the best researchers at analyzing data, and that 89% expect AI to match top researchers at scientific writing by 2030. These specific survey figures from the X post could not be verified against the retrieved Anthropic report, which does not include a full breakdown of survey responses for all tasks.

The 15-fold output increase is the high end of Anthropic's modeled scenario for certain fields, not a proven outcome Anthropic1. The report describes it as a hypothetical scenario.