AIWatch.

As AI becomes critical infrastructure for science, medicine, industry, and government, one question matters more than ever: who gets to understand how these systems work? At Ai2, that question is central to our work, and it’s what motivates our commitment to open science and open models. For us, openness is more than a licensing choice—it shapes how we build, release, and evaluate our work.

Open-weight models let people use and adapt AI for their own research and applications. They give organizations greater choice, reduce dependence on any single provider, and lower barriers to developing new applications and scientific tools.

With our fully open releases, we go further by sharing the training data, code, methods, checkpoints, evaluations, and documentation behind the models. We do this because we believe the ability to understand how AI works shouldn't be locked up in a few hands—it should be available to researchers and institutions working to advance scientific knowledge.

This is what open science looks like in AI: producing research that others can inspect, test, and extend.

Trust requires open evidence

Researchers need more than access to a model’s outputs. They need the data, code, checkpoints, evaluations, and training process behind it to reproduce results, investigate behavior, test limitations, and assess claims about safety and reliability. Trust comes from allowing independent researchers to examine the evidence rather than relying solely on a developer’s assurances.

The clearest measure of open research’s value is the new science it makes possible. By releasing the evidence behind our models, Ai2 enables independent researchers to examine questions that closed systems leave difficult – or impossible – to investigate.

For example, researchers at Northeastern and Johns Hopkins have used our Olmo models’ weights, internal representations, training data, and provenance to study demographic bias in clinical applications and whether models’ stated knowledge cutoffs match the information they actually rely on. Researchers at Arb Research and collaborating universities used Olmo 3’s training data and intermediate checkpoints to show that paraphrased copies of benchmark questions can inflate apparent model progress. Elsewhere, scientists at the University of Texas at Austin, Northeastern, and MD Anderson Cancer Center used Olmo 3 and infini-gram, Ai2’s engine for searching massive text corpora, to investigate how models reason about drug names.

In all of these cases, a fully open model was essential. Because the underlying evidence is available, other researchers can scrutinize these findings, reproduce the work, and extend it to new scientific and public-interest questions.

This matters as American researchers increasingly rely on capable open models developed abroad. A strong domestic base of open science gives U.S. institutions models they can inspect, adapt, and run themselves—supporting independent research at home while extending the reach of American science worldwide.

What is at stake for scientific progress

The next phase of U.S. AI policy will shape whether universities, nonprofit institutes, and public-interest researchers can continue to participate in advanced AI development. Open scientific models, datasets, research methods, and viable pathways for smaller research institutions to build advanced systems are important to continued scientific progress.

Maintaining access to open-source AI artifacts is essential to broad participation in AI research and development. Universities, nonprofit institutes, public-interest researchers, and smaller labs often rely on open models, training data, code, and research methods because they cannot reproduce them from scratch. Without that access, experimentation becomes harder, fewer research teams can contribute, and more of the field’s technical direction risks becoming concentrated inside a small number of companies.

U.S. scientific leadership has long depended on turning individual discoveries into foundations others can build on. In AI, open artifacts, shared infrastructure, and broad participation carry that tradition forward—expanding independent research, accelerating scientific progress, and enabling more people to understand, evaluate, and improve the systems shaping society.