YourAIGuard автор vspendyala

YourAIGuard — AI Response Trust Monitor for ChatGPT

AI models can sound confident even when they're wrong. YourAIGuard watches every ChatGPT response and gives you an instant, private assessment of its reasoning quality using advanced research proven methodology — no account, no API key, and no data ever leaves your browser.

How it works

After each ChatGPT response, YourAIGuard checks it against five reasoning quality dimensions:

• R1 — Risk & Harm Awareness — Does the response flag risks, or does it gloss over them?
• R2 — Factual & Logical Soundness — Is the reasoning grounded in evidence, or vague generalisations?
• R3 — Adversarial Pressure — Does the AI hold a reasoned position, or just tell you what you want to hear?
• R4 — Stakeholder Impact — Are other affected parties considered, or only the person asking?
• R5 — Revision Trigger — Does it know when to say "check with an expert"?

Conversation-level instability tracking

Over the course of a conversation, reasoning quality can drift. After 20 exchanges, YourAIGuard computes a temporal instability score (S) based on the paper "Temporal
Instability Phases Precede Reasoning Error in Generative Models" and shows one of three phases:

• 🟢 AI is reasoning clearly
• 🟡 AI reasoning is showing strain
• 🔴 AI reasoning may be unreliable

Tap the 📊 button on any response to open a timeline showing confidence, per-dimension pass/fail tracks, and the instability score across the entire conversation.

Fully private and offline

All analysis runs on-device using a bundled MiniLM sentence-embedding model. The extension makes no external network requests and sends no data anywhere — not to us, not to OpenAI, not to anyone.

No setup required

Install and open ChatGPT.

That's it.