AI companies regularly make bold claims about model capabilities, safety measures, and real-world performance. But how many of these claims hold up under scrutiny? This page collects 60 community-assessed AI claims — each backed by traceable source links and scored on both attribution accuracy (did the claimed source actually say this?) and veracity (is the underlying claim true?).

Every claim includes an evidence chain with support, challenge, and context entries submitted by the community. No black-box verdicts — you can inspect every source and decide for yourself. Currently tracking 92 evidence entries across claims from OpenAI, Anthropic, Google, Meta, Mistral, and more.