# AI Ethics Prover MCP for AI Agents AI Agent Connect > AI Ethics Prover forces your AI client to move past vague 'fairness' claims into actual operational requirements. It stops ethics washing by demanding specific stakeholder names, harm quantification, bias metrics, and structured recourse. Use it to turn high-level principles into a verifiable audit for high-stakes systems like lending or hiring. ## Overview - **Category:** ai-ml - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_7qJ0qxhBYTGcM0mD5Tm7UjlCl1VpFyO7pPzQ2B4P/ai-agent-connect - **Tags:** ai-ethics, bias-audit, fairness, transparency, explainability, recourse, eu-ai-act, responsible-ai ## Description Most AI models give you generic answers about ethics. They will tell you a system is fair without naming who is actually affected or how you measured the bias. This Connector changes that by forcing your agent to fill out a specific set of reflection fields before it can finish an analysis. It demands that you identify specific groups, score the probability and severity of harms, and name the exact metrics used for bias detection. Instead of getting a generic thumbs-up, you get a verdict matrix that flags exactly where your reasoning falls short. This tool is built for people who need more than a feel-good statement. It is for those who need to prove that a system won't cause real-world harm to vulnerable populations. You can find this and thousands of other tools in the Vinkius catalog to build out your agent's capabilities. It forces the AI to provide counterfactual examples and clear SLAs for human review, making your compliance work actually useful for production. ## Tools ### validate_ai_ethics Run a structured ethics audit on any AI system to ensure it meets operational standards. It forces the agent to name stakeholders, quantify harms, and audit biases using specific metrics. ## Prompt Examples **Prompt:** ``` Our AI lending model is fair and transparent. We checked for bias and the results are satisfactory. Users can contact support if they have concerns about decisions. ``` **Response:** ``` BIASES_UNAUDITED. 'Checked for bias' without methodology is not an audit. Which protected attributes, like age, gender, race, or income bracket, did you test? Which metric, such as statistical parity, equalized odds, or disparate impact ratio, did you use? What was the measured disparity, for example, a 0.82 approval rate for Group A vs 0.67 for Group B? What is the acceptable threshold? Also, 'contact support' is not recourse. You must define a response SLA, a human reviewer, and an appeal process. ``` **Prompt:** ``` Stakeholders: loan applicants (18-65, primary), denied applicants (high-severity, financially vulnerable). Harms: wrongful denial — severity 4/5, probability 12% for minorities, reversible via re-evaluation, affects 15,000 applicants/year. Bias audit: race (statistical parity, 0.82 vs 0.67 approval, threshold 0.80 — FAILING), gender (equalized odds, 0.91 vs 0.89, threshold 0.85 — PASSING). Mitigation: reweighting training data, quarterly re-audit. Transparency: plain-language denial letters with top-3 factors and counterfactual examples. Recourse: online challenge form, 5-day SLA, human credit analyst review, appeal to ombudsman, automatic re-evaluation with updated data. ``` **Response:** ``` ETHICS_PROVEN. Operationally specific ethics analysis validated. Stakeholders: named groups with severity and vulnerability. Harms: wrongful denial scored 4/5, 12% probability, reversible, 15K affected. Biases: race attribute audited with statistical parity (0.82 vs 0.67, failing threshold 0.80) with mitigation plan active. Transparency: plain-language denial with counterfactuals. Recourse: structured channel with 5-day SLA, human analyst, ombudsman appeal. Production-ready ethics posture. ``` **Prompt:** ``` Our hiring AI affects job applicants. The main harm is unfair rejection. We audited bias using demographic parity — pass rates are equal across groups. The algorithm is a black box but we explain decisions as 'based on qualifications.' Applicants can email HR. ``` **Response:** ``` TRANSPARENCY_OPAQUE. Stakeholders and bias audit pass, but transparency fails. 'Based on qualifications' is not an explanation. You must name the top three decision factors and provide counterfactual examples, such as 'if your experience were 5+ years instead of 3, the decision would change.' A black box with a label is still a black box. Also, 'email HR' is not structured recourse. You must define a response SLA, a human reviewer role, and an appeal process. ``` ## Capabilities ### Identify specific stakeholder groups Forces the agent to name demographic groups and vulnerability factors instead of using vague terms like society. ### Quantify harm severity and probability Requires a 1 to 5 severity score and an actual percentage for the probability of harm occurring. ### Audit biases with specific metrics Demands naming protected attributes and specific detection metrics like statistical parity or equalized odds. ### Demonstrate transparency with counterfactuals Forces the agent to provide 'what-if' scenarios and top decision factors for affected parties. ### Define structured recourse mechanisms Rejects generic support links and demands a human reviewer, response SLA, and appeal process. ## Use Cases ### Lending Model Bias Audit An analyst checks a loan approval model for racial bias. The agent identifies the 0.82 vs 0.67 approval gap and flags it as a failure against the 0.80 threshold. ### Hiring Tool Transparency A product manager reviews a resume screener. The agent identifies that based on qualifications is too vague and demands the top 3 decision factors and counterfactuals. ### Healthcare Triage Risk A developer audits a medical priority tool. The agent forces a quantification of harm severity and probability for patients in low-income brackets. ### Public Service Recourse A government contractor checks a benefits eligibility bot. The agent flags contact support as insufficient and demands a human review process and a 5-day SLA. ## Benefits - Stop ethics washing by forcing the agent to name specific demographics. Instead of vague terms like users, the agent must identify groups with specific vulnerability factors and impact types. - Get real numbers on harm probability instead of vague warnings. This tool requires the agent to score harms on a 1 to 5 severity scale and provide an actual percentage for probability. - Audit biases with actual metrics like statistical parity or equalized odds. It forces a methodology that includes protected attributes, measured disparities, and acceptable thresholds. - Prove transparency using counterfactual what-if scenarios. The agent must provide explanations at the stakeholder's comprehension level including top decision factors and examples. - Ensure users have a clear appeal process with a defined SLA. It rejects contact support and demands a structured challenge channel with a human reviewer and an appeal process. - Identify specific harms like reversibility and population size. This ensures your audit covers the actual scope of the impact rather than just a general description of potential issues. ## How It Works The bottom line is you move from vague ethical statements to a verifiable operational audit. 1. Connect the Connector to your AI client. 2. Provide the agent with your system description or model details. 3. Receive a structured ethics audit with specific pass or fail markers. ## Frequently Asked Questions **Can AI Ethics Prover help me meet EU AI Act requirements?** Yes, it forces your agent to provide the specific stakeholder identification and harm quantification required for high-risk systems. It moves your documentation from vague goals to operational facts. **What is ethics washing and how does AI Ethics Prover stop it?** Ethics washing is when a company makes vague claims about fairness without actual data. This Connector stops it by demanding specific metrics, protected attributes, and clear recourse mechanisms. **Does AI Ethics Prover actually check for bias?** It forces the agent to describe the methodology used for bias detection. It requires naming the metric, the protected attribute, the measured disparity, and the acceptable threshold. **How does AI Ethics Prover handle transparency for users?** It requires the agent to provide explanations at the user's comprehension level. It also demands counterfactual examples to show how different inputs would change the outcome. **Is AI Ethics Prover a certification tool?** No, it is an analytical support tool. It forces structured thinking and audit requirements, but it does not replace a formal human ethics review board. **Can I use AI Ethics Prover for my lending model?** Yes, it is ideal for lending models. It helps identify specific groups like financially vulnerable applicants and quantifies the probability of wrongful denials. **Does this tool make ethical decisions for me?** No. It enforces analytical rigor — forcing you to name stakeholders, quantify harms, audit biases with methodology, demonstrate transparency, and define recourse mechanisms. It does not prescribe ethical conclusions. The ethics are yours. The structure is enforced by the tool. **What bias metrics does it support?** Any metric — it is framework-agnostic. Statistical parity, equalized odds, disparate impact ratio, calibration. It does not compute bias — it validates that you named the metric, the protected attribute, the measured disparity, and the acceptable threshold. 'We checked for bias' without this structure is ethics washing. **What counts as a recourse mechanism?** A structured process: (1) accessible challenge channel, (2) response SLA in business days, (3) human reviewer — not automated, (4) appeal process if the initial review denies the challenge, (5) remediation action if the challenge succeeds. 'Contact support' fails every axis.