Of course, just because research is published in a journal rather than a preprint server doesn’t mean it’s inherently risk-free. But it does mean that any obvious dangers are more likely to be discovered during the review process. “In fact, the key difference between journals and preprint servers is the depth of review, and the journal publication process may be more likely to identify risks,” Smith said.

The risks of open publishing do not end with biological research. In artificial intelligence, a similar movement to openly share code and data means there is potential for abuse. In November 2019, OpenAI announced that it would not publicly release the full contents of its new language model, GPT-2, which can generate text and answer questions independently, because of concerns about “malicious applications of the technology” — meaning it has the potential to spread Fake news and disinformation. Instead, OpenAI would release a much smaller version of the model for researchers to modify, a decision that drew criticism at the time. (It went on to release full models in November of that year.) Its successor, GPT-3, was released in 2020 and was found to be capable of writing child pornography.

The two largest preprint servers, medRxiv, founded in 2019 for publishing medical research, and bioRxiv, which was founded in 2013 for biological research, publicly state on their website that they check for “dual-use research of concern” not published on them on the website. website. “All manuscripts were submitted screened for plagiarism, non-scientific content, inappropriate article types, and material that could endanger individual patients or public health,” a statement on medRxiv said. “The latter may include, but are not limited to, descriptions of Dual-use research and work that challenges or may compromise accepted public health measures and recommendations regarding the transmission, immunization, and treatment of infectious diseases.”

BioRxiv risk has been an issue since bioRxiv’s inception, said Richard Sever, one of bioRxiv’s co-founders and assistant director of Cold Spring Harbor Laboratory Press. (Sever was a peer reviewer of Smith and Sandbrink’s paper.) He joked that in the early days of arXiv, a physical science preprint server launched in 1991, there were concerns about nuclear weapons. Today’s bioRxiv worries about bioweapons.

Sever estimates that bioRxiv and medRxiv receive about 200 submissions per day, each of which is watched by more than one pair of eyes. They got “a lot of crap” and were thrown away immediately, but the rest of the submissions went into a pool to be sifted by practicing scientists. If someone flagged a paper of potential concern during the initial screening process, it was passed on to the management team for consideration before a final decision. “We always try to err on the side of caution,” Sever said. He believes that, so far, nothing has proven to be dangerous.

Several papers have been rejected over the years because the team believes they fall into the dual-use research category of interest. When the pandemic hit, the issue became even more pressing. By April 2021, the two servers had posted more than 15,000 preprints on Covid-19. It turned into an internal debate: whether the high risk of a pandemic meant they were morally obliged to publish papers on what they called “pandemic pathogens” — such as Sars-CoV-2 — that they used to Perhaps traditionally rejected? “There’s a change in how risk-benefit calculations are done,” Sever said.