📄 Papers from a single fake research network have reached clinical guidelines, policy documents and patents. Federica Silvi and Leslie McIntosh of Digital Science traced 1,975 papers whose authors listed the Pharmakon Neuroscience Network, an authorship-for-sale operation exposed in 2022, and followed where they were cited. Of those papers, 12 were cited in clinical guidelines, 57 in policy documents and 480 in patents. Of the 278 researchers involved, 90% kept publishing after the exposé. The analysis · Retraction Watch

➗ OpenAI has pulled its sponsorship of Caltech's Mathathon, Gizmodo reports, after an open letter drew 771 signatories from Caltech and the wider mathematical research community. The letter said 40 hours is not enough for real mathematics and that the verification burden would land on the community. The event still runs on 30 October. This is the same fight over credit that led last week's issue. The open letter · Gizmodo

🏛️ Researchers are suing the NIH over what their complaint calls "an ideological litmus test for NIH-funded research". The challenge, filed last week, says a text-analysis tool screened grants against a list that held at least 235 flagged terms in February, and that hundreds of grants were cut. One plaintiff says she was asked to describe gender equity research without using the words gender, equity or sexual. Inside Higher Ed

🏅 A 2015 paper by a Nobel laureate has a permanent expression of concern. Retraction Watch reports the notice and says the same researcher tops its list of Nobel winners by retractions, with 15. An expression of concern is not a retraction; it is the journal telling you to read with care while questions stay open. Retraction Watch

🪧 The Lancet's staff strike on Tuesday. Members of the National Union of Journalists at The Lancet, about 110 staff across several of its 30 journals, will strike for one day on 22 September and then work to rule from 23 to 30 September. The union asked for 7.5% in the first year; Elsevier countered with 4.1% for the first year and 2.9% for the second.

Why it matters: The union says it is the first strike in the journal's 200-year history. Editorial staff are the people who chase your reviewers, answer your status email and move your revision along. A week of contractual duties only is a week when nothing informal happens. Informal is how manuscripts move.

What I would do: If you have a manuscript with a Lancet journal this week, do not read silence as a decision. If a revision deadline falls inside the strike and the work-to-rule that follows it, email the editorial office now and ask for a written extension, rather than assuming one.

Reference: Retraction Watch, 16 Sep 2026

Coverage: National Union of Journalists · 2026-09-16

🎭 Science pulled a story after finding its author was not who they said. Science removed a news article after determining the author was impersonating the identity of an established science journalist, with its own news editors and with sources in the story. Two quotes in the piece were later found in stories published by other writers.

Why it matters: The forgery was not a fake paper or a fake reviewer. It was a fake reporter, and it cleared a professional news desk. Publishing has spent years learning to verify reviewers. Nobody verifies the reporter. The cost landed on the scientists who answered in good faith and are now quoted in a story that no longer exists.

What I would do: When a journalist you do not know asks for an interview, spend two minutes before you answer: find them on the outlet's own staff or contributor page, and reply to the address listed there rather than the one that contacted you. If you cannot find them, ask the outlet's newsroom to confirm the commission.

Reference: Retraction Watch, 19 Sep 2026

Coverage: Science, editor's note, 16 Sep 2026 · 2026-09-16

🤖 A tool that turns a paper into an agent you can question. A Nature paper published on 16 September describes Paper2Agent, which reads a paper's text, code and data and builds an agent that answers questions about the work. On the AlphaGenome paper it produced an agent in about 45 minutes for $14 of compute. Across 100 computational biology papers it succeeded on 74.

Why it matters: Look at the failures rather than the headline. The ones that did not work were mostly papers whose code or data was incomplete, which is exactly what stops a human trying to reproduce you, minus the politeness. Your repository is about to be read by software that never emails to ask which script runs first. It does not ask twice.

What I would do: Open your last paper's repository and try to run it from the README alone, in a clean folder. Fix what stops you: a missing environment file, a hard-coded path, a data file that lives only on your laptop. That is the difference between being reusable and being cited once.

Reference: Nature, 16 Sep 2026

Coverage: Nature news, 16 Sep 2026 · 2026-09-16

Open the manuscript you are closest to submitting and write two things. First, a short log: every AI tool you used on it, what it touched (literature search, drafting, translation, editing, code, figures) and what you changed afterwards. Second, the sentence you would put in the paper. Then check it against the journal's policy, because they differ: some want a statement only for drafting and editing, others want every use. If your log is empty, write that you used none, and you are done. The point is to write it now, while you still remember, instead of reconstructing it under an editor's deadline.