What's this about?
Commercial AI products run on hidden system prompts — developer instructions that govern the model's behavior but are almost never disclosed to users or regulators. AISPA proposes user-centric auditing: ways to probe what those hidden instructions make an application do.
What exactly is a "system prompt"?
It's the standing instruction sheet a developer gives the model before your conversation starts — persona, rules, forbidden topics, sales incentives. You never see it, but it shapes every answer. Auditing from the outside means inferring the rules from behavior, like testing a bank's policies by making transactions (16 upvotes).
Why it matters
As AI products mediate advice and decisions, "what were they instructed to do?" becomes a consumer-protection question — this gives regulators and users a methodology.