OpenAI's 'Shepherd' Strategy Sparks Backlash: Researchers Demand Open Access to AI Breakthroughs

In a significant development that has reignited tensions within the global scientific community, OpenAI has announced impressive breakthroughs in mathematics, computer science, and physics, introducing a controversial new protocol for disseminating these findings. The company stated that it is collaborating with expert 'shepherds' to determine the optimal strategy for sharing these results. However, this approach has drawn sharp criticism from researchers who argue that such gatekeeping undermines the fundamental principles of open science and hinders the rapid verification of AI-generated discoveries. The recent announcement follows a turbulent period marked by the sudden release of hundreds of preprints in early October, which overwhelmed the academic community. This time, OpenAI has adopted a tiered release strategy: the mathematical results have been made public, allowing researchers to organize their own explanatory efforts; a computer science finding is being prepared for release by an external group of researchers; and the physics results remain withheld, with the company deferring to external experts on how to proceed. Despite the varying levels of disclosure, experts who spoke to The Verge described the results as 'huge' and 'major breakthroughs,' acknowledging their potential to solve some of the most consequential unsolved problems in their respective fields. However, the excitement over the scientific merit is tempered by deep frustration regarding OpenAI's conduct. Constantin Kogler, a researcher at the Institute for Advanced Study, welcomed the manageable batch of three results but questioned the logic of restricting initial access to a handpicked group of distinguished scientists. 'It doesn't make any sense that first only a few distinguished scientists can look at them,' Kogler stated, emphasizing that public announcement should inherently mean availability for broader scrutiny. Similarly, Andreas Karch, a physics professor at the University of Texas at Austin, labeled the handpicking of the expert group a 'disaster,' arguing that large AI labs already possess immense power and should not be selecting 'winners and losers' in the scientific discourse. The criticism is further amplified by the recent stance of the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), which warned against such practices just a day prior to the announcement. AGMAI argued that selecting specific individuals to receive output first is 'unnecessary' and risks biased selection and unequal access. Instead, the group recommended that AI labs release outputs on scholarly repositories with formalized proofs and provide funding to support the community in understanding these results. Following the latest release, AGMAI urged OpenAI to make public the natural-language and Lean outputs behind its physics announcements without delay, noting that the formalization is complete and there is no scientific reason for further delay. The debate highlights a growing rift between AI developers and the traditional academic establishment. While OpenAI aims to manage the narrative and ensure the results are 'meaningful and understandable,' the scientific community insists that democratizing access is essential for robust peer review and the advancement of knowledge. As Hassett noted, making findings openly available could lead to better science and encourage wider adoption of new tools. The resolution of this impasse will likely set a precedent for how future AI-driven scientific discoveries are shared and validated in the coming years.