ai safety

A Bash-first playbook for Linux AI incident response: why AI failures are unique (state beyond code, nondeterminism, cross-system boundaries, safety), a ready toolchain and triage script to freeze/collect evidence, methods to trace network/files/GPU, redacted prompt/output logging, a root-cause taxonomy, a real PII-leak case study, plus one-liners and a checklist to deploy, pin, audit, and rehearse for faster, reproducible fixes.