RoleReveal
A consumer-facing 'show your work' browser extension that surfaces the hidden system prompt, guardrails, and tool list of any AI assistant on the page — so anyone can see what rules their chatbot, copilot, or AI tutor is actually operating under.
Non-technical professionals (lawyers, teachers, recruiters, HR staff) using AI tools at work and wondering what the tool is and isn't allowed to do
- One-click inspection of any web AI: shows system prompt length, persona, refusal rules, and tool access in plain English
- 'What would you say if I asked X?' probe button that hints at hidden guardrails without jailbreaking
- Side-by-side comparison of two AI tools' policies (e.g., ChatGPT vs. Copilot for HR use)
- Export a one-page transparency PDF to share with your IT or compliance team
The leaked-prompts trend proves there's mass public interest in what's inside AI products — most non-engineers can't read or find system prompts themselves, so a consumer browser layer unlocks transparency for a much bigger audience than the GitHub repo.
Public interest in leaked system prompts is real and large — asgeirtj/system_prompts_leaks has ~41,890 stars / 6,936 forks and a competing 30-tool repo hit 136K stars, with Medium + Reddit threads widely read. But the audience is overwhelmingly developers and AI hobbyists, not the non-technical lawyers/teachers/HR target.System Prompts Leaks Reference Collection | OpenTools ↗GitHub repo exposes system prompts from 28+ AI coding tools: what it reveals ↗
No consumer-facing browser extension that automatically surfaces a chatbot's system prompt/guardrails appears to exist; closest are security-focused 'prompt injection detector' extensions (e.g., AI Shield) and static GitHub collections. Wide-open space for a polished consumer product.AI Shield — Hidden Prompt Injection Detector - GitHub ↗Malicious AI Assistant Extensions Harvest LLM Chat Histories ↗
Hard to monetize — all leading artifacts (GitHub leaks, Medium guides, Reddit extraction prompts) are free; no evidence of paid competitors or willingness-to-pay for transparency. Non-technical professionals are unlikely to pay subscription for something they could ask about indirectly, and enterprises already get this via vendor compliance docs.GitHub - asgeirtj/system_prompts_leaks ↗How to Reveal the Hidden Rules Behind ChatGPT's System Prompts in One Quick Step ↗
AI vendors are actively hardening against extraction (jailbreak-resistant system prompts, server-side enforcement), and the 'leaked prompts' trend could compress quickly as architectures change. Conversely, regulatory pressure (EU AI Act, transparency mandates) could keep demand alive — net mixed/uncertain.How to Extract System Instructions from Any LLM (Yes, Even ChatGPT, Claude, Gemini, Grok, etc.) ↗
Fundamentally harder than it looks: system prompts are server-side and not exposed in the DOM of ChatGPT/Claude/Gemini web apps, so the extension can only ship a curated DB of known-leaked prompts or rely on prompt-injection tricks that vendors actively patch. This makes reliable coverage across 'any AI on the page' very difficult without ongoing manual curation.Top AI Tools Exposed to Man in the Prompt Browser Extension Attack ↗Extracting instruction prompts from custom GPTs ↗