{
  "name": "Human-Centered Artificial Intelligence Framework (HAIAI)",
  "shortName": "HAIAI",
  "description": "Human-Centered Artificial Intelligence (HAI) is a research institute and approach to AI development that prioritizes the needs of human users. HAI aims to create AI systems that are ethical, understandable, and designed to enhance human capabilities rather than replace them.",
  "iconUrl": "https://hai.stanford.edu/sites/default/files/news/teaser-images/hai_thumbnail_0.png",
  "stages": [
    {
      "stageName": "Fairness",
      "systemComponents": [
        {
          "componentName": "Data Collection and Processing",
          "risks": [
            {
              "riskId": "1.1",
              "title": "Demographic Bias in Data",
              "definition": "Design a data collection pipeline that samples from a broad and diverse set of demographics. Implement validation checks to ensure the data's demographic representativeness aligns with expectations.",
              "addressedByControls": ["HAIAI-1", "HAIAI-2", "HAIAI-3"]
            },
            {
              "riskId": "1.2",
              "title": "Lack of Transparency in Model Development",
              "definition": "Create a documentation process that describes your model's architecture, data sources, and training methodology. Ensure this documentation is easily accessible to authorized third parties for audits and reviews.",
              "addressedByControls": [
                "HAIAI-4",
                "HAIAI-5",
                "HAIAI-6",
                "HAIAI-7"
              ]
            },
            {
              "riskId": "1.3",
              "title": "Lack of Diversity in Stakeholder Engagement",
              "definition": "Develop a stakeholder engagement framework that includes representatives from various demographics during model design, development, and review phases. Keep a record of feedback and actions taken based on stakeholder input.",
              "addressedByControls": [
                "HAIAI-8",
                "HAIAI-9",
                "HAIAI-10",
                "HAIAI-11"
              ]
            },
            {
              "riskId": "1.4",
              "title": "Lack of Demographic Testing and Bias Detection",
              "definition": "Implement automated testing scripts to evaluate model performance across key demographic segments. Use statistical analysis to detect discrepancies or biases and document results in an accessible report.",
              "addressedByControls": ["HAIAI-12", "HAIAI-13", "HAIAI-14"]
            },
            {
              "riskId": "1.5",
              "title": "Bias Mitigation Techniques Integration",
              "definition": "Integrate bias mitigation algorithms or techniques within your model development process. Use methods like reweighting, adversarial debiasing, or data augmentation to address known biases. Test and document the effectiveness of these techniques in reducing bias.",
              "addressedByControls": ["HAIAI-15"]
            }
          ]
        }
      ]
    },
    {
      "stageName": "Data Governance",
      "systemComponents": [
        {
          "componentName": "Data Management System",
          "risks": [
            {
              "riskId": "2.1",
              "title": "Non-Compliance and Consent Management",
              "definition": "Develop a compliance and consent management system that includes regular audits and adherence to standards, ensuring data use aligns with all applicable laws and regulations.",
              "addressedByControls": ["HAIAI-16", "HAIAI-17"]
            },
            {
              "riskId": "2.2",
              "title": "Data Quality Assurance",
              "definition": "Implement a data quality assurance framework that assesses and ensures completeness, uniqueness, consistency, and accuracy in data collection and preparation.",
              "addressedByControls": [
                "HAIAI-18",
                "HAIAI-19",
                "HAIAI-20",
                "HAIAI-21"
              ]
            },
            {
              "riskId": "2.3",
              "title": "Demographic Representativeness",
              "definition": "Implement robust validation checks to ensure that the data is representative and accurately reflects the demographic context in which the final model or system will be used.",
              "addressedByControls": ["HAIAI-22"]
            },
            {
              "riskId": "2.4",
              "title": "Continuous Data Audits and Updates",
              "definition": "Establish continuous, systematic data audits and updates to maintain and ensure the ongoing relevancy and accuracy of the data.",
              "addressedByControls": [
                "HAIAI-23",
                "HAIAI-24",
                "HAIAI-25",
                "HAIAI-26"
              ]
            },
            {
              "riskId": "2.5",
              "title": "Dataset Documentation and Traceability",
              "definition": "Implement a detailed dataset documentation and traceability process to maintain comprehensive records and accountability throughout the AI life cycle. Some potential documentations include: data origin, collection process, variables used, accessing of dataset, consent and compliance forms, relevance to system.",
              "addressedByControls": ["HAIAI-27", "HAIAI-28"]
            },
            {
              "riskId": "2.6",
              "title": "Data Remediation Strategies",
              "definition": "Develop comprehensive remediation strategies (Error correction, missing data handling, imbalanced data) and maintain detailed documentation for datasets with shortcomings, outlining specific corrective actions, responsible parties, and implementation timelines to address data quality issues and ensure ongoing data integrity.",
              "addressedByControls": ["HAIAI-29", "HAIAI-30"]
            }
          ]
        }
      ]
    },
    {
      "stageName": "Transparency and Explainability",
      "systemComponents": [
        {
          "componentName": "Documentation Process",
          "risks": [
            {
              "riskId": "3.1",
              "title": "Inadequate Documentation Process",
              "definition": "Implement a comprehensive documentation process for AI development that records algorithm design choices, data sources, intended use cases, and identified limitations of the model to ensure transparency and accountability throughout the development lifecycle.",
              "addressedByControls": [
                "HAIAI-31",
                "HAIAI-32",
                "HAIAI-33",
                "HAIAI-34"
              ]
            }
          ]
        },
        {
          "componentName": "Training Programs",
          "risks": [
            {
              "riskId": "3.2",
              "title": "Lack of Comprehensive Training Programs",
              "definition": "Develop and implement comprehensive training programs for stakeholders and users to educate them about the intended use cases, capabilities, and limitations of the model, ensuring proper understanding about the model and its uses.",
              "addressedByControls": [
                "HAIAI-35",
                "HAIAI-36",
                "HAIAI-37",
                "HAIAI-38",
                "HAIAI-39",
                "HAIAI-40"
              ]
            },
            {
              "riskId": "3.3",
              "title": "Inadequate Model Selection Strategy",
              "definition": "Adopt a strategy prioritizing simpler, highly interpretable models in scenarios where transparency and understandability are crucial, even with trade-offs in performance to allow for easy explanation.",
              "addressedByControls": ["HAIAI-41", "HAIAI-42"]
            },
            {
              "riskId": "3.4",
              "title": "Lack of Model Explainability Tools",
              "definition": "Implement model explainability tools (saliency maps, LIME, SHAP, PDP) to provide clear visual explanations of how models make decisions, enhancing transparency and facilitating deeper understanding of model behavior",
              "addressedByControls": ["HAIAI-43"]
            }
          ]
        }
      ]
    },
    {
      "stageName": "Reliability",
      "systemComponents": [
        {
          "componentName": "Model Reliability",
          "risks": [
            {
              "riskId": "4.1",
              "title": "Model Error Mitigation",
              "definition": "Implement mitigation measures for model errors and handling low-confidence output.",
              "addressedByControls": ["HAIAI-44", "HAIAI-45"]
            },
            {
              "riskId": "4.2",
              "title": "Failover Plan Development",
              "definition": "Develop a failover plan (i.e., redundancy measures) to ensure system or model availability, enhancing reliability and minimizing downtime in critical applications.",
              "addressedByControls": ["HAIAI-46"]
            },
            {
              "riskId": "4.3",
              "title": "Rigorous Testing Protocols",
              "definition": "Implement rigorous testing protocols such as red team exercises to evaluate models and systems for vulnerabilities or harmful behaviors.",
              "addressedByControls": ["HAIAI-47", "HAIAI-48"]
            },
            {
              "riskId": "4.4",
              "title": "Robust Security Measures",
              "definition": "Implement robust security measures to prevent adversarial attacks (input validation, anomaly detection, security audits, adversarial training techniques).",
              "addressedByControls": ["HAIAI-49"]
            },
            {
              "riskId": "4.5",
              "title": "Confidence Scoring Mechanisms",
              "definition": "Implement confidence scoring mechanisms for model outputs.",
              "addressedByControls": ["HAIAI-50"]
            },
            {
              "riskId": "4.6",
              "title": "Comprehensive Test Case Development",
              "definition": "Develop comprehensive test cases that encompass a wide range of scenarios to ensure security and reliability among a wide variety of scenarios and use cases.",
              "addressedByControls": ["HAIAI-51"]
            }
          ]
        }
      ]
    },
    {
      "stageName": "Security",
      "systemComponents": [
        {
          "componentName": "AI Security Measures",
          "risks": [
            {
              "riskId": "5.1",
              "title": "Basic Cybersecurity Hygiene Practices Implementation",
              "definition": "Implement basic cybersecurity hygiene practices (multifactor authentication, strict access controls, ongoing employee training)",
              "addressedByControls": ["HAIAI-52", "HAIAI-53"]
            },
            {
              "riskId": "5.2",
              "title": "Third-Party Vendor Cybersecurity Vetting Process Establishment",
              "definition": "Implement a rigorous vetting and validation process for the cybersecurity measures of third-party vendors in the supply chain if using third parties for monitoring tasks.",
              "addressedByControls": ["HAIAI-54", "HAIAI-55"]
            },
            {
              "riskId": "5.3",
              "title": "Dedicated AI Cybersecurity Team Establishment",
              "definition": "Establish and implement a dedicated AI cybersecurity team or train existing personnel specifically for AI-specific cybersecurity challenges.",
              "addressedByControls": ["HAIAI-56", "HAIAI-57"]
            },
            {
              "riskId": "5.4",
              "title": "Technical AI-Specific Cybersecurity Measures Implementation",
              "definition": "Implement technical AI-specific cybersecurity checks and measures (adversarial testing to evaluate robustness, comprehensive vulnerability assessments to identify potential security flaws, data security measures to protect against unauthorized access and data breaches).",
              "addressedByControls": ["HAIAI-58", "HAIAI-59"]
            },
            {
              "riskId": "5.5",
              "title": "Resource Allocation for AI-Specific Cybersecurity Research and Monitoring",
              "definition": "Allocate resources for ongoing research and monitoring of evolving AI-specific cybersecurity risks; make sure they are integrated into existing cybersecurity processes.",
              "addressedByControls": ["HAIAI-60", "HAIAI-61"]
            }
          ]
        }
      ]
    }
  ],
  "controls": [
    {
      "controlId": "HAIAI-1",
      "title": "Diverse Demographic Sampling",
      "description": "Developers of AI models must design a data collection pipeline that samples from diverse demographics.",
      "controlCategory": "Curation",
      "readableControlId": "HAIAI 1",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Inclusivity in Data Sources",
          "description": "Ensure data is collected from sources representing a wide range of demographic groups to avoid bias."
        }
      ]
    },
    {
      "controlId": "HAIAI-2",
      "title": "Data Validation Checks",
      "description": "Developers of AI models must implement data validation checks.",
      "controlCategory": "Curation",
      "readableControlId": "HAIAI 2",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Accuracy and Consistency Verification",
          "description": "Regularly check data for accuracy and consistency to ensure it meets demographic representativeness standards."
        }
      ]
    },
    {
      "controlId": "HAIAI-3",
      "title": "Ensuring Demographic Representativeness",
      "description": "Developers of AI models must ensure that the demographic representativeness of the data is sufficiently diverse.",
      "controlCategory": "Curation",
      "readableControlId": "HAIAI 3",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Diversity Metrics Monitoring",
          "description": "Monitor and report on diversity metrics regularly to ensure the data reflects the required demographic spread."
        }
      ]
    },
    {
      "controlId": "HAIAI-4",
      "title": "Model Architecture Documentation",
      "description": "Developers of AI models must create documentation that describes the model's architecture.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 4",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Detailed Architecture Description",
          "description": "Provide a detailed description of the model architecture, including layers, activation functions, and hyperparameters."
        }
      ]
    },
    {
      "controlId": "HAIAI-5",
      "title": "Data Source Documentation",
      "description": "Developers of AI models must create documentation that describes the model's data sources.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 5",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Source and Nature of Data",
          "description": "Include information on the origin, nature, and preprocessing of data sources."
        }
      ]
    },
    {
      "controlId": "HAIAI-6",
      "title": "Training Methodology Documentation",
      "description": "Developers of AI models must create documentation that describes the model's training methodology.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 6",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Process Details",
          "description": "Document the steps taken during training, including data splitting, augmentation, and model evaluation."
        }
      ]
    },
    {
      "controlId": "HAIAI-7",
      "title": "Accessibility of Documentation",
      "description": "Developers of AI models must ensure that the documentation of data is easily accessible to third-party auditors.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 7",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Access Protocols",
          "description": "Establish protocols for granting access to documentation to authorized personnel."
        }
      ]
    },
    {
      "controlId": "HAIAI-8",
      "title": "Diverse Stakeholder Engagement Framework (Design Phase)",
      "description": "Developers of AI models must develop a stakeholder engagement framework that includes representatives from different demographics during the model design phase.",
      "controlCategory": "Design",
      "readableControlId": "HAIAI 8",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Inclusive Design Process",
          "description": "Ensure that stakeholders from diverse demographics are included in the design phase to capture a wide range of perspectives."
        }
      ]
    },
    {
      "controlId": "HAIAI-9",
      "title": "Diverse Stakeholder Engagement Framework (Development Phase)",
      "description": "Developers of AI models must develop a stakeholder engagement framework that includes representatives from various demographics during the model development phase.",
      "controlCategory": "Architecture",
      "readableControlId": "HAIAI 9",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Inclusive Development Process",
          "description": "Include stakeholders from diverse demographics during the development phase to ensure the model meets the needs of different user groups."
        }
      ]
    },
    {
      "controlId": "HAIAI-10",
      "title": "Diverse Stakeholder Engagement Framework (Review Phase)",
      "description": "Developers of AI models must develop a stakeholder engagement framework that includes representatives from various demographics during the model review phase.",
      "controlCategory": "Review",
      "readableControlId": "HAIAI 10",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Inclusive Review Process",
          "description": "Engage stakeholders from diverse demographics during the review phase to gather feedback and ensure the model's effectiveness and inclusivity."
        }
      ]
    },
    {
      "controlId": "HAIAI-11",
      "title": "Feedback Documentation",
      "description": "Developers of AI models must keep a record of feedback based on stakeholder input.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 11",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Record Keeping",
          "description": "Maintain detailed records of stakeholder feedback, including the source, date, and actions taken."
        }
      ]
    },
    {
      "controlId": "HAIAI-12",
      "title": "Automated Testing Scripts Implementation",
      "description": "Developers of AI models must implement automated testing scripts to evaluate model performance across key demographic segments.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 12",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Script Development",
          "description": "Develop automated testing scripts tailored to evaluate model performance across different demographic segments."
        }
      ]
    },
    {
      "controlId": "HAIAI-13",
      "title": "Statistical Bias Analysis",
      "description": "Developers of AI models must use statistical analysis to detect discrepancies in an accessible report.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 13",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Discrepancy Detection",
          "description": "Apply statistical methods to identify any discrepancies in model performance across demographic groups."
        }
      ]
    },
    {
      "controlId": "HAIAI-14",
      "title": "Bias Detection Analysis",
      "description": "Developers of AI models must use statistical analysis to detect biases in an accessible report.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 14",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Bias Detection Methods",
          "description": "Utilize statistical methods to identify and document any biases present in the model's performance across demographic segments."
        }
      ]
    },
    {
      "controlId": "HAIAI-15",
      "title": "Integration of Bias Mitigation Techniques",
      "description": "Developers of AI models must integrate bias mitigation algorithms within the development process.",
      "controlCategory": "Development",
      "readableControlId": "HAIAI 15",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Technique Integration",
          "description": "Ensure that bias mitigation techniques are effectively integrated into the model development process to address known biases."
        },
        {
          "criteriaId": "2",
          "title": "Effectiveness Testing",
          "description": "Test and document the effectiveness of bias mitigation techniques in reducing bias within the model."
        }
      ]
    },
    {
      "controlId": "HAIAI-16",
      "title": "Compliance and Consent Management System",
      "description": "Developers of AI models must develop a compliance and consent management system that includes regular audits.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 16",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Regular Audits",
          "description": "Conduct regular audits to ensure compliance with established consent management protocols."
        }
      ]
    },
    {
      "controlId": "HAIAI-17",
      "title": "Legal and Regulatory Alignment",
      "description": "Developers of AI models must ensure that data use aligns with all applicable laws and regulations.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 17",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Legal Compliance",
          "description": "Verify that data usage policies comply with relevant legal and regulatory frameworks."
        }
      ]
    },
    {
      "controlId": "HAIAI-18",
      "title": "Data Completeness Assurance",
      "description": "Developers of AI models must implement a data quality assurance framework that ensures completeness in data collection and preparation.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 18",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Completeness Verification",
          "description": "Ensure that all necessary data points are collected and available for analysis."
        }
      ]
    },
    {
      "controlId": "HAIAI-19",
      "title": "Data Uniqueness Assurance",
      "description": "Developers of AI models must implement a data quality assurance framework that ensures uniqueness in data collection and preparation.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 19",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Uniqueness Verification",
          "description": "Ensure that data records are unique and free of duplicates."
        }
      ]
    },
    {
      "controlId": "HAIAI-20",
      "title": "Data Consistency Assurance",
      "description": "Developers of AI models must implement a data quality assurance framework that ensures consistency in data collection and preparation.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 20",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Consistency Verification",
          "description": "Ensure that data is consistent across different sources and throughout the data lifecycle."
        }
      ]
    },
    {
      "controlId": "HAIAI-21",
      "title": "Data Accuracy Assurance",
      "description": "Developers of AI models must implement a data quality assurance framework that ensures accuracy in data collection and preparation.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 21",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Accuracy Verification",
          "description": "Ensure that data accurately represents real-world conditions and values."
        }
      ]
    },
    {
      "controlId": "HAIAI-22",
      "title": "Robust Data Validation",
      "description": "Developers of AI models must implement robust validation checks to ensure that the data is representative and accurately reflects the demographic context in which the final model or system will be used.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 22",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Demographic Representation Check",
          "description": "Ensure data validation processes include checks for demographic representativeness."
        }
      ]
    },
    {
      "controlId": "HAIAI-23",
      "title": "Systematic Data Audits for Relevancy",
      "description": "Developers of AI models must establish systematic data audits, ensuring the ongoing relevancy of the data.",
      "controlCategory": "Evaluation/Curation",
      "readableControlId": "HAIAI 23",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Relevancy Audit Procedure",
          "description": "Implement regular audits to ensure data remains relevant to current application contexts."
        }
      ]
    },
    {
      "controlId": "HAIAI-24",
      "title": "Systematic Data Audits for Accuracy",
      "description": "Developers of AI models must establish systematic data audits, ensuring the ongoing accuracy of the data.",
      "controlCategory": "Evaluation/Curation",
      "readableControlId": "HAIAI 24",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Accuracy Audit Procedure",
          "description": "Conduct regular audits to verify and maintain the accuracy of the data used."
        }
      ]
    },
    {
      "controlId": "HAIAI-25",
      "title": "Systematic Data Updates for Relevancy",
      "description": "Developers of AI models must establish systematic data updates, ensuring the ongoing relevancy of the data.",
      "controlCategory": "Evaluation/Curation",
      "readableControlId": "HAIAI 25",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Relevancy Update Procedure",
          "description": "Implement procedures for updating data to maintain relevancy as context and applications evolve."
        }
      ]
    },
    {
      "controlId": "HAIAI-26",
      "title": "Systematic Data Updates for Accuracy",
      "description": "Developers of AI models must establish systematic data updates, ensuring the ongoing accuracy of the data.",
      "controlCategory": "Evaluation/Curation",
      "readableControlId": "HAIAI 26",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Accuracy Update Procedure",
          "description": "Ensure regular updates to data are conducted to maintain and verify its accuracy."
        }
      ]
    },
    {
      "controlId": "HAIAI-27",
      "title": "Detailed Dataset Documentation",
      "description": "Developers of AI models must implement a detailed dataset documentation process to maintain comprehensive records throughout the AI life cycle.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 27",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Comprehensive Dataset Records",
          "description": "Maintain detailed documentation of data origin, collection process, variables used, dataset access, consent and compliance forms, and relevance to the system."
        }
      ]
    },
    {
      "controlId": "HAIAI-28",
      "title": "Detailed Dataset Traceability",
      "description": "Developers of AI models must implement a detailed dataset traceability process to maintain comprehensive records throughout the AI life cycle.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 28",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Traceability Records",
          "description": "Ensure traceability of the dataset  by documenting the data flow, modifications, and usage across different stages of the AI life cycle."
        }
      ]
    },
    {
      "controlId": "HAIAI-29",
      "title": "Comprehensive Remediation Strategies",
      "description": "Developers of AI models must develop comprehensive remediation strategies for datasets with shortcomings.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 29",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Remediation Strategies for Data Issues",
          "description": "Implement strategies such as error correction, handling missing data, and addressing imbalanced data to ensure data integrity."
        }
      ]
    },
    {
      "controlId": "HAIAI-30",
      "title": "Detailed Documentation for Dataset Shortcomings",
      "description": "Developers of AI models must maintain detailed documentation for datasets with shortcomings.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 30",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Documentation of Corrective Actions",
          "description": "Document specific corrective actions, responsible parties, and implementation timelines for addressing data quality issues."
        }
      ]
    },
    {
      "controlId": "HAIAI-31",
      "title": "Algorithm Design Choices Documentation",
      "description": "Developers of AI models must implement a comprehensive documentation process for AI development that records algorithm design choices for the model.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 31",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Documentation Process",
          "description": "A comprehensive documentation process is implemented to record algorithm design choices."
        }
      ]
    },
    {
      "controlId": "HAIAI-32",
      "title": "Data Sources Documentation",
      "description": "Developers of AI models must implement a comprehensive documentation process for AI development that records data and sources of the model.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 32",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Documentation Process",
          "description": "A comprehensive documentation process is implemented to record data sources."
        }
      ]
    },
    {
      "controlId": "HAIAI-33",
      "title": "Intended Use Cases Documentation",
      "description": "Developers of AI models must implement a comprehensive documentation process for AI development that records intended use cases of the model.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 33",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Documentation Process",
          "description": "A comprehensive documentation process is implemented to record intended use cases."
        }
      ]
    },
    {
      "controlId": "HAIAI-34",
      "title": "Limitations Documentation",
      "description": "Developers of AI models must implement a comprehensive documentation process for AI development that records identified limitations of the model.",
      "controlCategory": "Documentation",
      "readableControlId": "HAIAI 34",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Documentation Process",
          "description": "A comprehensive documentation process is implemented to record identified limitations."
        }
      ]
    },
    {
      "controlId": "HAIAI-35",
      "title": "Intended Use Cases Training",
      "description": "Developers of AI models must develop comprehensive training programs for stakeholders and users to educate them about the intended use cases of the model.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 35",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Program",
          "description": "Comprehensive training programs are developed for stakeholders and users to educate them about intended use cases."
        }
      ]
    },
    {
      "controlId": "HAIAI-36",
      "title": "Model Capabilities Training",
      "description": "Developers of AI models must develop comprehensive training programs for stakeholders and users to educate them about the capabilities of the model.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 36",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Program",
          "description": "Comprehensive training programs are developed for stakeholders and users to educate them about model capabilities."
        }
      ]
    },
    {
      "controlId": "HAIAI-37",
      "title": "Model Limitations Training",
      "description": "Developers of AI models must develop comprehensive training programs for stakeholders and users to educate them about the limitations of the model.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 37",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Program",
          "description": "Comprehensive training programs are developed for stakeholders and users to educate them about model limitations."
        }
      ]
    },
    {
      "controlId": "HAIAI-38",
      "title": "Intended Use Cases Training (Implementation)",
      "description": "Developers of AI models must implement comprehensive training programs for stakeholders and users to educate them about the intended use cases of the model.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 38",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Program Implementation",
          "description": "Comprehensive training programs are implemented for stakeholders and users to educate them about intended use cases."
        }
      ]
    },
    {
      "controlId": "HAIAI-39",
      "title": "Model Capabilities Training (Implementation)",
      "description": "Developers of AI models must implement comprehensive training programs for stakeholders and users to educate them about the capabilities of the model.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 39",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Program Implementation",
          "description": "Comprehensive training programs are implemented for stakeholders and users to educate them about model capabilities."
        }
      ]
    },
    {
      "controlId": "HAIAI-40",
      "title": "Model Limitations Training (Implementation)",
      "description": "Developers of AI models must implement comprehensive training programs for stakeholders and users to educate them about the limitations of the model.",
      "controlCategory": "Feedback",
      "readableControlId": "HAIAI 40",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Training Program Implementation",
          "description": "Comprehensive training programs are implemented for stakeholders and users to educate them about model limitations."
        }
      ]
    },
    {
      "controlId": "HAIAI-41",
      "title": "Prioritize Simpler Models",
      "description": "Developers of AI models must adopt a strategy prioritizing simpler models.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 41",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Model Strategy",
          "description": "Developers adopt a strategy prioritizing simpler models."
        }
      ]
    },
    {
      "controlId": "HAIAI-42",
      "title": "Prioritize Interpretable Models",
      "description": "Developers of AI models must adopt a strategy prioritizing highly interpretable models.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 42",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Model Strategy",
          "description": "Developers adopt a strategy prioritizing highly interpretable models."
        }
      ]
    },
    {
      "controlId": "HAIAI-43",
      "title": "Implement Model Explainability Tools",
      "description": "Developers of AI models must implement model explainability tools to provide clear visual explanations of how models make decisions.",
      "controlCategory": "Participation",
      "readableControlId": "HAIAI 43",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Tool Implementation",
          "description": "Developers implement model explainability tools to provide visual explanations of model decisions."
        }
      ]
    },
    {
      "controlId": "HAIAI-44",
      "title": "Model Error Mitigation",
      "description": "Developers of AI models must implement mitigation measures for model errors.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 44",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Mitigation Measures",
          "description": "Mitigation measures for model errors must be clearly defined and implemented to address potential issues."
        }
      ]
    },
    {
      "controlId": "HAIAI-45",
      "title": "Handling Low-Confidence Output",
      "description": "Developers of AI models must implement mitigation measures for handling low-confidence output.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 45",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Mitigation Measures",
          "description": "Mitigation measures for handling low-confidence output must be clearly defined and implemented to address potential issues."
        }
      ]
    },
    {
      "controlId": "HAIAI-46",
      "title": "Failover Plan Development",
      "description": "Developers of AI models must develop a failover plan to ensure model availability.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 46",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Failover Plan Implementation",
          "description": "Developers must create and implement a failover plan that outlines procedures for maintaining model availability in the event of system failures or disruptions."
        }
      ]
    },
    {
      "controlId": "HAIAI-47",
      "title": "Rigorous Testing Protocol for Vulnerabilities",
      "description": "Developers of AI models must implement rigorous testing protocols to evaluate models for vulnerabilities.",
      "controlCategory": "Testing",
      "readableControlId": "HAIAI 47",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Vulnerability Testing",
          "description": "Developers should conduct thorough testing procedures to identify and address vulnerabilities in AI models, ensuring robustness against potential exploits or security breaches."
        }
      ]
    },
    {
      "controlId": "HAIAI-48",
      "title": "Rigorous Testing Protocol for Harmful Behaviors",
      "description": "Developers of AI models must implement rigorous testing protocols to evaluate models for harmful behaviors.",
      "controlCategory": "Testing",
      "readableControlId": "HAIAI 48",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Behavior Testing",
          "description": "Developers should conduct comprehensive testing to identify and mitigate any harmful behaviors exhibited by AI models, ensuring ethical and responsible AI deployment."
        }
      ]
    },
    {
      "controlId": "HAIAI-49",
      "title": "Adversarial Attack Prevention Measures",
      "description": "Developers of AI models must implement security measures to prevent adversarial attacks.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 49",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Security Implementation",
          "description": "Developers should incorporate robust security measures into AI models to mitigate the risk of adversarial attacks, ensuring the integrity and reliability of the system."
        }
      ]
    },
    {
      "controlId": "HAIAI-50",
      "title": "Confidence Scoring Mechanisms Implementation",
      "description": "Developers of AI models must implement confidence scoring mechanisms for model outputs.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 50",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Confidence Assessment",
          "description": "Developers should incorporate confidence scoring mechanisms into AI models to provide insights into the reliability and certainty of model predictions, aiding users in assessing the trustworthiness of model outputs."
        }
      ]
    },
    {
      "controlId": "HAIAI-51",
      "title": "Comprehensive Test Case Development",
      "description": "Developers of AI models must develop comprehensive test cases to ensure security among a wide variety of scenarios.",
      "controlCategory": "Protection",
      "readableControlId": "HAIAI 51",
      "severity": "high",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Test Case Coverage",
          "description": "Developers should create thorough test cases that cover a diverse range of scenarios to assess the security and reliability of AI models across various use cases, ensuring robustness and resilience against potential vulnerabilities."
        }
      ]
    },
    {
      "controlId": "HAIAI-52",
      "title": "Basic Cybersecurity Hygiene Practices Implementation",
      "description": "Developers of AI models must implement basic cybersecurity hygiene practices.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 52",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Implementation of Basic Cybersecurity Practices",
          "description": "The AI model development process incorporates multifactor authentication, strict access controls, and ongoing employee training for cybersecurity."
        }
      ]
    },
    {
      "controlId": "HAIAI-53",
      "title": "Ongoing Employee Security Training",
      "description": "Developers of AI models must ensure ongoing employee security training.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 53",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Ongoing Employee Training",
          "description": "Regular training sessions are conducted to educate employees on security best practices, including threat awareness and response."
        }
      ]
    },
    {
      "controlId": "HAIAI-54",
      "title": "Third-Party Vendor Cybersecurity Vetting Process Establishment",
      "description": "Developers of AI models must implement a rigorous vetting process for the cybersecurity measures of third-party vendors in the supply chain.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 54",
      "severity": "low",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Cybersecurity Vetting Process",
          "description": "A thorough evaluation process is established to assess the cybersecurity measures implemented by third-party vendors, ensuring alignment with organizational security standards."
        }
      ]
    },
    {
      "controlId": "HAIAI-55",
      "title": "Third-Party Vendor Cybersecurity Validation Process Establishment",
      "description": "Developers of AI models must implement a rigorous validation process for the cybersecurity measures of third-party vendors in the supply chain.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 55",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Cybersecurity Validation Process",
          "description": "A validation procedure is established to verify the effectiveness and reliability of the cybersecurity measures implemented by third-party vendors."
        }
      ]
    },
    {
      "controlId": "HAIAI-56",
      "title": "Dedicated AI Cybersecurity Team Establishment",
      "description": "Developers of AI models must establish a dedicated AI cybersecurity team.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 56",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Establishment of a Dedicated Team",
          "description": "A specialized team focused on AI-specific cybersecurity challenges is formed, comprising skilled professionals with relevant expertise."
        }
      ]
    },
    {
      "controlId": "HAIAI-57",
      "title": "AI-Specific Cybersecurity Training for Existing Personnel",
      "description": "Developers of AI models must train existing personnel specifically for AI-specific cybersecurity challenges.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 57",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Personnel Training",
          "description": "Existing personnel receive specialized training sessions tailored to address AI-specific cybersecurity risks and mitigation strategies."
        }
      ]
    },
    {
      "controlId": "HAIAI-58",
      "title": "Technical AI-Specific Cybersecurity Checks Implementation",
      "description": "Developers of AI models must implement technical AI-specific cybersecurity checks.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 58",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Implementation of Cybersecurity Checks",
          "description": "Technical measures, including regular scans and assessments, are implemented to identify and address AI-specific cybersecurity vulnerabilities."
        }
      ]
    },
    {
      "controlId": "HAIAI-59",
      "title": "Technical AI-Specific Cybersecurity Measures Implementation",
      "description": "Developers of AI models must implement technical AI-specific cybersecurity measures.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 59",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Implementation of Cybersecurity Measures",
          "description": "Technical safeguards, such as encryption, access controls, and intrusion detection systems, are deployed to protect AI systems from cyber threats."
        }
      ]
    },
    {
      "controlId": "HAIAI-60",
      "title": "Resource Allocation for AI-Specific Cybersecurity Research",
      "description": "Developers of AI models must allocate resources for ongoing research on evolving AI-specific cybersecurity risks.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 60",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Resource Allocation for Research",
          "description": "Sufficient resources, including funding, personnel, and time, are allocated to conduct ongoing research on emerging AI-specific cybersecurity threats and vulnerabilities."
        }
      ]
    },
    {
      "controlId": "HAIAI-61",
      "title": "Resource Allocation for AI-Specific Cybersecurity Monitoring",
      "description": "Developers of AI models must allocate resources for ongoing monitoring of evolving AI-specific cybersecurity risks.",
      "controlCategory": "Evaluation",
      "readableControlId": "HAIAI 61",
      "severity": "medium",
      "automationPlatforms": [],
      "criteria": [
        {
          "criteriaId": "1",
          "title": "Resource Allocation for Monitoring",
          "description": "Adequate resources, such as tools, personnel, and infrastructure, are dedicated to continuously monitoring AI systems for emerging cybersecurity risks and threats."
        }
      ]
    }
  ]
}