Ask HN: Are guardrails being baked into models, or enforced outside them?

By arkj · 2026-09-03 · 2 points · 0 comments

Anyone else feeling like Claude got more defensive since the latest upgrade? A question for people who know: is this kind of caution trained into the weights, or is it a policy/guardrail layer sitting outside the model?

Open the full discussion on BetterNews