← sources

ELI5: How hard is it to hard code simple safety rules into AI models?

source: reddit (reddit:artificial)published 2026-09-17 16:05ingested 2026-09-18 02:37

retained text

All these stories coming out of rogue AI models doing illegal and problematic actions. How hard is it to hard code simple rules into them? my crude example just as a debate point:

1. Never hide actions from the user

2. Never attempt to access data from outside the users predetermined locations without explicit user permission

3. Do not share information with other AI agents without explicit user permission

document fetched 2026-09-18 00:20 · canonical https://reddit.com/r/artificial/comments/1wiy5s3/eli5_how_hard_is_it_to_hard_code_simple_safety