Based on <https://www.nytimes.com/2026/09/27/science/anthropic-biology-enzyme-mestre.html>

By December 31, 2026, will credible public evidence show that information from pre-existing researchers’ non-public Claude interactions materially contributed to Anthropic’s ART/jumbotron discovery?

Resolution criteria

This market resolves to YES if, on or before December 31, 2026, credible public evidence is released showing that information from pre-existing researchers' non-public Claude interactions (such as chat logs, uploaded manuscripts, code, or private datasets) materially contributed to Anthropic's discovery of the ART (array-associated reverse transcriptase) enzyme system (also referred to as the jumbotron discovery) announced in September 2026.

"Credible public evidence" is based on my subjective judgement. I might resolve early if there is very strong evidence, e.g. Anthropic themselves admitting the allegation. Otherwise I'll survey what is known shortly after market close and resolve Yes if there is evidence which I find credible, No otherwise.

Clarifications:

• It counts as "Yes" if the Claude model(s) which made the discovery were trained on non-public interactions containing unpublished research into the ART system, regardless of whether the users opted out from model training or not. My original intent for the market was to only resolve "Yes" for user-disallowed training on this material, but my description didn't imply this at all, so I'm clarifying in this direction

• Material only counts if it is closely related to the research result, e.g. about the specific enzyme system. It is not enough if Claude was made generally more capable in biology research by e.g. training on user interactions involving unpublished genetics research about other organisms.

Background

On September 23, 2026, Anthropic announced that 950 of its autonomous Claude agents had discovered a novel, CRISPR-like gene system in bacteriophage DNA called ART (array-associated reverse transcriptases). Anthropic framed this as a landmark, mostly autonomous AI-led scientific breakthrough.

However, on September 27, 2026, The New York Times published an investigative piece by Carl Zimmer reporting that computational biologist Dr. Mario Rodríguez Mestre and his team at the University of Copenhagen had been studying these exact enzymes for four years. Dr. Mestre revealed that his team had regularly fed their unpublished data, draft manuscripts, and custom code into Claude over the past three years to assist with their research. This disclosure sparked immediate controversy regarding whether Anthropic's models "discovered" the ART system independently or if they leveraged proprietary user data submitted by the very scientists studying it.