A hot LocalLLaMA thread points to Anthropic confirming Claude Code limits aimed at game cloning, with users reporting Opus 5.5 also blocking reverse-engineering-style asks. The OP’s case for local/open models isn’t only privacy or price: provider policy can suddenly fence off legitimate adjacent work (security research, interoperability, debugging closed formats). Community replies usually split between “guardrails are fine” and “blanket refusals are why we keep weights at home.”Practical note: keep a local open-weight coding model for policy-sensitive or offline workflows, and treat cloud assistants as best-effort—not the only tool that can open a binary or map a protocol. Always stay inside the law; self-hosting is about control and continuity, not a free pass for piracy.Has a cloud model refused a legitimate reverse-engineering / security task you needed? What local stack did you fall back to?Source: https://www.reddit.com/r/LocalLLaMA/comments/1xm33gr/yet_another_reason_to_self_host_anthropic_has/