Is the published literature still reliable?

The base rate of unsound published work is climbing. Retractions passed a record 10,000 in a single year. Paper mills, operations that sell fabricated or manipulated manuscripts, are growing far faster than legitimate research, with fraudulent output estimated to double on a timescale of months rather than the ~15 years for genuine literature (Richardson et al. 2025). Fabricated references, where papers cite studies that were never published, have risen sharply (Topaz et al. 2026). The tools many teams now lean on, including general AI assistants, can't reliably tell sound work from manufactured work. The warning signs are usually in the paper itself: methods that don't match the claims, conclusions the design can't support, citations to evidence that doesn't exist. Reading for those signals is the defence.