Unbiased AI-powered news
An audit of nearly 2.5 million biomedical papers identified more than 4,000 fabricated references across nearly 3,000 studies. The rate of papers containing at least one fake reference increased from one in 2,828 in 2023 to one in 277 in early 2026.
rediff.comA Columbia University researcher who studies artificial intelligence nearly published a paper containing a fabricated reference generated by an AI tool. The journal returned the manuscript with questions about the citation, prompting an investigation into how often such errors appear in published work.
5 million biomedical papers and 97 million citations indexed on PubMed Central. The study, published earlier this month in The Lancet, found more than 4,000 fabricated references across nearly 3,000 papers.
The rate of papers containing at least one fabricated reference grew more than 12-fold over three years. In 2023 the rate stood at one in 2,828 papers; by 2025 it had reached one in 458. Topaz said the increase in fabricated references went “vertical” in 2024, coinciding with wider use of AI tools in research. The study noted that not all fabricated references were necessarily AI-generated.
Topaz described the chain of citations that links individual studies to systematic reviews and clinical guidelines. Fabricated references inserted at any point can be carried forward into later work that informs medical decisions. 4 percent of papers containing fabricated references had not been retracted at the time of the study.
Verification practices vary across journals, and no standard mechanism exists to screen the citation record retroactively. Topaz stated that the solution lies in adding verification steps to existing workflows rather than stopping the use of AI tools.
He noted that the problem involves unverified AI output entering the permanent scientific record.
nypost.comSuper PACs tied to Anthropic and OpenAI have spent more than $37 million on congressional primaries this cycle. The groups have outspent candidates in some races and focused on candidates who back differing approaches to AI regulation.
flipboard.comPresident Trump met Anthropic CEO Dario Amodei at the G7 summit and described talks on restoring access to Fable 5 and Mythos 5 as progressing. The company disabled the models for all users after an administration order to block foreign nationals.
Al JazeeraThe U.S. directed Anthropic to block all foreign nationals from its two frontier AI models last week. Anthropic took the systems offline; G7 allies discussed a trusted-partner access plan.