Leading AI Labs Have Largely Stopped Publishing Research, Science Analysis Finds

A new analysis published in Science confirms a trend the research community has long observed: the leading AI startups — OpenAI, Anthropic, and others — have largely stopped publishing peer-reviewed research. Where these companies once shared foundational work openly, they now treat novel techniques as proprietary assets, citing competitive pressure and safety concerns as justifications.

The shift is a reversal of the field's early culture. The most consequential AI research of the past decade moved through public channels — conferences, preprints, journals — and that openness accelerated progress across the entire ecosystem. The analysis in Science finds that access for independent researchers has contracted as the gap between the frontier and publicly available work has widened.

The word the analysis uses for where the field is heading is "balkanizing" — a field once held together by shared knowledge fragmenting into competing institutional silos where critical advances don't cross organizational lines. Whether competitive pressure or safety is the dominant driver, both rationales point to the same outcome.

The study doesn't specify what would reverse the trend. But the documentation of it — in Science, with named companies — puts the question squarely on the table for policymakers and funders who assumed openness would remain the default.

Read the full story at Science