• METRICS
  • IMPACT
  • LOG
  • HUMAN

How to quantify AI model harms with your AI

Connect your AI to the Ethics Prover to turn qualitative risk concerns into a structured dataset of measurable harms.

Ask AI about this page

Short answer

How can I measure AI model harms?

Your AI client uses the Ethics Prover to map abstract risk descriptions to specific numerical values for severity, probability, and population impact. This process converts subjective concerns into a structured risk profile. You get a clear breakdown of these metrics for every identified harm.

The output data.

Where your risk data lands.

The AI processes your input into these four distinct categories.

METRICS

Numerical risk scores

The AI calculates specific values for severity and probability. You get a quantitative baseline for every risk identified.

IMPACT

Population reach

The tool estimates how many users or specific groups are affected. This helps you see the scale of a potential harm.

LOG

Structured risk registry

All findings are organized into a clean format. You can move these directly into your internal safety documentation.

HUMAN

Expert review needed

High-severity or low-confidence scores are flagged for you. You decide if the AI's quantification matches your internal standards.

The workflow

What your AI does when a risk statement arrives.

The AI handles the heavy lifting of translation and calculation.

  1. Parse risk text

    Your AI reads your raw notes or model documentation to find hidden ethical risks.

    validate_ai_ethics

  2. Assign scores

    The AI maps identified harms to specific severity and probability scales.

    validate_ai_ethics

  3. Calculate impact

    The tool determines the scope of the harm across different user populations.

    validate_ai_ethics

  4. Flag outliers

    The AI highlights risks that require a human safety engineer to review.

    validate_ai_ethics

Try it

Copy these to start.

Use these prompts to kick off the quantification process.

Starting points

Replace the bracketed text with your specific model details or risk descriptions.

AI Ethics Prover Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_7qJ0qxhBYTGcM0mD5Tm7UjlCl1VpFyO7pPzQ2B4P/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — AI Ethics Prover capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "ai-ethics-prover-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_7qJ0qxhBYTGcM0mD5Tm7UjlCl1VpFyO7pPzQ2B4P/mcp"
    }
  }
}
Copy into chat04

  • Analyze this model description for potential harms: [paste model doc]

  • Quantify the severity and probability for this risk: [paste risk statement]

  • Look at my safety report in [paste file name] and extract the population impact for each harm.

  • Review the risks listed in this GitHub repo [paste repo link] and give me a structured table of scores.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect AI Ethics Prover once, then ask.

Just link the MCP to your preferred client. Your credentials stay encrypted, and you can start asking questions immediately.

Connect AI Ethics Prover to your AI

FAQ

How this task behaves.

  • 01

    Can the AI automatically fix the harms it finds?

    No. The AI only quantifies and identifies the risks. You must decide how to mitigate them.

  • 02

    Does the AI change my original risk documents?

    No. The AI reads your data to perform calculations but does not write back to or modify your source files.

  • 03

    How accurate are the severity scores?

    The scores are mathematical estimates based on your input. They are meant to guide your review, not replace professional judgment.

  • 04

    Can I use this with any AI client?

    Yes, as long as your client supports the Model Context Protocol (MCP).

  • 05

    What happens if the AI is unsure about a risk?

    The AI flags low-confidence findings so you know which results need a human eyes-on check.

  • More questions about AI Ethics Prover? The Connector page answers them. See everything the AI Ethics Prover Connector can do

Connect AI Ethics Prover to Claude, Cursor, ChatGPT & more