AI policy

Deterministic first. Evidence shown. Humans accountable.

FilthyGM uses governed AI only inside explicit capabilities with fixed releases, product-scoped tools, budgets, validation, and kill switches.

01

Allowed uses

AI may help parse a rumour into a draft structure, explain an already-calculated analysis, assist research and claim extraction, translate or narrate approved content, cluster moderation queues, and draft private fantasy explanations. It does not create canonical hockey facts or silently settle receipts.

02

Release control

Each release pins provider model, prompt, tool schema, retrieval, rubric, policy, generation settings, locale and tone rules, budget, evaluation evidence, approvals, traffic allocation, rollback target, and kill-switch state.

  • Prompt-injection and data-exfiltration tests
  • Unsupported-claim and temporal-leakage tests
  • French and English quality tests
  • Safety, dignity, privacy, and cost thresholds
03

Disclosure and failure

Private automated analysis says Generated by FilthyGM AI from the evidence shown and lists its release. Material AI assistance on staff work is disclosed; the named author and approving editor remain accountable. If validation or a dependency fails, generation is unavailable—never replaced with invented output.