Use this thread for consequential AI claims that need checking: capability announcements, synthetic media, labor claims, safety claims, benchmark results, and platform policy changes.
Link the original claim and the strongest independent evidence you can find. Note the publication date and product version. AI systems change quickly, and an accurate old claim can become a misleading current one.
Example:
A company announces its new model “passes the bar exam with a score in the 90th percentile.” The benchmark result is real. What is not stated: the exam questions were part of the training data, the model was tested on a leaked version, or the score was achieved after dozens of attempts with cherry-picked results. The claim is technically true. The implication, that the model reasons like a lawyer, is the work the framing does. A benchmark score is not evidence of understanding. It is evidence of pattern matching on a specific test.