AI article
We Tested Our Own AI Safety Checks. They Caught Nothing.
Community description: Our model invented 4 of 12 answers when the evidence was removed. Behind our own safety checks it invented 6, and the checks refused 0 of 24. They verify where words come from, not whether they answer
Dev.to | Oct 7, 2026 | DEVALAND
Automated excerpt
Quick answer: we tested the four safety checks in our own open-source Romanian question-answering demo against a model that was allowed to invent answers. C, numbers: does every number in the answer appear in the cited source? D, grounding: does every sentence share enough content words with one cited source?
Selected automatically from source text; not independently written or fact-checked. Read the original for full context.