Ethical Safeguards (FIBMAP)

Fairness, Inclusiveness, Bias Mitigation, Accountability, Privacy

Ethical Safeguards set the compass for responsible AI.

It encompasses five drivers: Fairness, Inclusiveness, Bias Mitigation, Accountability, and Privacy. They clarify what the system should protect, who it should serve, and how responsibilities are assigned before design or code begins. define the populations and contexts they intend to serve, measure and explain differential outcomes, and document choices that affect people and data. Clear ownership and auditable processes make responsibility real in daily work. Privacy by design limits collection and use to what is lawful and necessary, with strong controls and honest disclosure. Taken as a whole, Ethical Safeguards keep the system aligned with human values while providing practical guidance that teams can follow at every stage of the lifecycle.

Ethical Safeguards's 5 Drivers

FIBMAP

Fairness

Fairness keeps outcomes equitable across people and contexts. It focuses on clarifying target populations, testing for disparate impact, and documenting trade-offs when performance differs by subgroup. It prompts teams to choose appropriate fairness notions for the task, measure them transparently, and explain residual gaps. Fairness is not a single number. It is an explicit, auditable stance about how benefits and burdens are distributed.

Inclusiveness

Inclusiveness ensures people can access, understand, and influence the system regardless of background or ability. It emphasizes representative participation in requirements, language and accessibility best practices in UX, and feedback channels that surface overlooked needs. Inclusivity treats users and affected stakeholders as co-designers, not edge cases, broadening who the system serves and reducing exclusion that can compound downstream harms.

Bias Mitigation

Bias mitigation addresses skew introduced by data, modeling choices, and operations. It promotes careful dataset curation, traceable preprocessing, appropriate controls during training and evaluation, and runtime checks that catch drift or proxy effects. The goal is not to pretend bias vanishes, but to surface where it can arise, apply proportionate safeguards, and document residual risks so decisions remain transparent and correctable.

Accountability

Accountability makes responsibility concrete and enforceable. It defines owners for requirements, data, models, deployments, and monitoring, each with clear duties, sign-offs, and escalation paths. It favors auditable processes, versioned artifacts, and explanations that allow independent review. When issues occur, accountability enables timely remediation, learning, and communication with stakeholders, turning governance from policy text into day-to-day practice.

Privacy

Privacy protects individuals’ data and expectations across the lifecycle. It stresses data minimization, lawful and purposeful use, strong security controls, and user-respecting choices such as consent, transparency, and deletion. Technical measures (e.g., de-identification, access controls) pair with organizational safeguards and clear disclosures. Privacy treats personal information as a duty of care, preventing misuse while still enabling legitimate, proportionate value.

Ethical Safeguards Inter-Driver Relationship List

The following table summarizes the 60 ethical safeguards related, inter-driver relationships. The full 105 relationships can be viewed here:

Note: The convention when displaying drivers D_s vs. D_t, is to display the first driver alphabetically as D_s.

Drivers	Relationship	Explanation	Example
Inter-Pillar Relationships
Pillar: Ethical Safeguards
Accountability vs. Bias Mitigation	■ Reinforcing	Accountability involves ensuring AI systems do not cause harm through biases, closely aligning with bias mitigation efforts to ensure fairness (Ferrara, 2024 )	Regular bias audits reinforce accountability in AI hiring systems, reducing discriminatory outcomes and enhancing fairness (Cheong, 2024 )
Accountability vs. Fairness	■ Reinforcing	Accountability requires fairness to ensure equitable AI outcomes, linking the two drivers for ethical AI system design (Ferrara, 2024 )	In financial AI systems, fairness audits enhance accountability by preventing discriminatory lending practices and ensuring equitable treatment (Saura & Debasa, 2022 )
Accountability vs. Inclusiveness	■ Reinforcing	Accountability ensures inclusiveness by requiring equitable representation and consideration of diverse perspectives in AI system governance (Leslie, 2019 )	A company implements inclusive decision-making practices, reinforcing accountability by reducing disparities in AI outcomes (Bullock et al., 2024 )
Accountability vs. Privacy	■ Tensioned	Accountability can conflict with privacy, as complete transparency might infringe on data protection norms (Solove, 2025 )	Implementing exhaustive audit trails ensures accountability but could compromise individuals’ privacy in sensitive sectors (Solove, 2025 )
Bias Mitigation vs. Fairness	■ Tensioned	While both aim to reduce injustices, techniques for fairness (e.g., demographic parity) can sometimes contradict bias mitigation goals (Ferrara, 2024 )	Ensuring demographic parity in hiring algorithms might lead to the over-representation of certain groups, raising concerns about individual fairness (Dubber et al., 2020 )
Bias Mitigation vs. Inclusiveness	■ Reinforcing	Bias mitigation supports inclusiveness by actively addressing representation gaps, enhancing fairness across AI applications (Ferrara, 2024 )	Implementing bias audits ensures diverse datasets in educational AI models, promoting inclusiveness (Ferrara, 2024 )
Bias Mitigation vs. Privacy	■ Tensioned	Bias mitigation can conflict with privacy when data diversity requires sensitive personal information (Ferrara, 2024 )	Healthcare AI often struggles to balance privacy laws with the need for diverse training data (Ferrara, 2024 )
Fairness vs. Inclusiveness	■ Reinforcing	Inclusiveness enhances fairness by broadening AI’s scope to reflect diverse societal elements equitably (Shams et al., 2023 )	Inclusive AI hiring prevents gender disparity by reflecting diversity through fair data representation (Shams et al., 2023 )
Fairness vs. Privacy	■ Tensioned	Tensions arise as fairness needs ample data, potentially conflicting with privacy expectations (Cheong, 2024 )	Fair lending AI seeks demographic data for fairness, challenging privacy rights (Cheong, 2024 )
Inclusiveness vs. Privacy	■ Tensioned	Privacy-preserving techniques can limit data diversity, compromising inclusiveness (d’Aliberti et al., 2024 )	Differential privacy in healthcare AI might obscure patterns relevant to minority groups (d’Aliberti et al., 2024 )
Cross-Pillar Relationships
Pillar: Ethical Safeguards vs. Operational Integrity
Accountability vs. Explainability	■ Reinforcing	Accountability promotes explainability by requiring justifications for AI decisions, fostering transparency and informed oversight (Busuioc, 2021 )	Implementing clear explanations in credit scoring ensures accountability and compliance with regulations, enhancing stakeholder trust (Cheong, 2024 )
Accountability vs. Governance	■ Reinforcing	Governance frameworks incorporate accountability to ensure decisions in AI systems are monitored, traceable, and responsibly managed (Bullock et al., 2024 )	An AI firm employs governance logs to ensure accountable decision-making, facilitating regulatory compliance and stakeholder trust (Dubber et al., 2020 )
Accountability vs. Interpretability	■ Reinforcing	Accountability and interpretability enhance transparency and trust, essential for effective AI system governance (Dubber et al., 2020 )	In finance, regulators use interpretable AI to ensure banks’ accountability by tracking decisions (Ananny & Crawford, 2018 )
Accountability vs. Robustness	■ Tensioned	Accountability’s demand for traceability may compromise robustness by exposing sensitive vulnerabilities (Leslie, 2019 )	In autonomous vehicles, robust privacy measures can conflict with traceability required for accountability (Busuioc, 2021 )
Accountability vs. Safety	■ Reinforcing	Accountability ensures AI safety by demanding human oversight and system verification, enhancing procedural safeguards (Leslie, 2019 )	In autonomous vehicles, accountability through rigorous standards enhances safety measures ensuring fail-safe operations (Leslie, 2019 )
Accountability vs. Security	■ Reinforcing	Accountability enhances security by ensuring responsible data management and risk identification (Voeneky et al., 2022 )	Regular audits on AI systems’ security protocols ensure accountability and safety for data governance (Voeneky et al., 2022 )
Bias Mitigation vs. Explainability	■ Tensioned	Bias mitigation can obscure model operations, conflicting with the transparency needed for explainability (Rudin, 2019 )	In high-stakes justice applications, improving model explainability can compromise bias mitigation efforts (Busuioc, 2021 )
Bias Mitigation vs. Governance	■ Reinforcing	Governance frameworks regularly incorporate bias mitigation strategies, reinforcing ethical AI implementation (Ferrara, 2024 )	AI governance policies in finance often include bias audits, ensuring ethical compliance (Ferrara, 2024 )
Bias Mitigation vs. Interpretability	■ Reinforcing	Interpretability aids bias detection, supporting equitable AI systems by elucidating model decisions (Ferrara, 2024 )	Interpretable healthcare models reveal biases in diagnostic outputs, promoting fair treatment (Ferrara, 2024 )
Bias Mitigation vs. Robustness	■ Reinforcing	Bias mitigation enhances robustness by incorporating diverse data, reducing systematic vulnerabilities in AI models (Ferrara, 2024 )	Inclusive datasets in AI model training improve both bias mitigation and system robustness (Ferrara, 2024 )
Bias Mitigation vs. Safety	■ Reinforcing	Bias mitigation increases safety by addressing discrimination risks, central to safe AI deployment (Ferrara, 2024 )	Ensuring fair training data mitigates bias-related risks in AI models, enhancing safety in autonomous vehicles (Ferrara, 2024 )
Bias Mitigation vs. Security	■ Reinforcing	Bias mitigation enhances security by reducing vulnerabilities that arise from discriminatory models (Habbal et al., 2024 )	Including bias audits in AI-driven fraud detection systems strengthens security protocols (Habbal et al., 2024 )
Explainability vs. Fairness	■ Reinforcing	Explainability assists in ensuring fairness by elucidating biases, enabling equitable AI systems (Ferrara, 2024 )	In credit scoring, explainable models help identify discrimination, promoting fairer lending practices (Ferrara, 2024 )
Explainability vs. Inclusiveness	■ Reinforcing	Explainability promotes inclusiveness by making AI decisions understandable, encouraging equitable stakeholder participation (Shams et al., 2023 )	Explainable AI models help identify underrepresented groups’ needs, ensuring inclusive design in public policy (Shams et al., 2023 )
Explainability vs. Privacy	■ Tensioned	Explainability can jeopardize privacy by revealing sensitive algorithm details (Solove, 2025 )	Disclosing algorithm logic in healthcare AI might infringe patient data privacy (Solove, 2025 )
Fairness vs. Governance	■ Reinforcing	Governance ensures fairness by establishing regulatory frameworks that guide AI systems towards unbiased practices (Cath, 2018 )	The EU AI Act mandates fairness algorithms under governance to prevent discrimination in employment (Cath, 2018 )
Fairness vs. Interpretability	■ Reinforcing	Interpretability fosters fairness by making opaque AI systems comprehensible, allowing equitable scrutiny and accountability (Binns, 2018 )	Interpretable algorithms in credit scoring identify biases, supporting fairness standards and promoting equitable lending (Bateni et al., 2022 )
Fairness vs. Robustness	■ Tensioned	Fairness might necessitate modifications that decrease robustness (Tocchetti et al., 2022 )	Adjustments to AI models for fairness in loan approvals might reduce performance across datasets (Braiek & Khomh, 2024 )
Fairness vs. Safety	■ Tensioned	Fairness can conflict with safety since safety may require restrictive measures that impact equitable access (Leslie, 2019 )	Self-driving algorithms balanced between passenger safety and fair pedestrian detection can lead to safety and fairness trade-offs (Cath, 2018 )
Fairness vs. Security	■ Tensioned	Fairness needs data transparency, often conflicting with strict security protocols prohibiting data access (Leslie et al., 2024 )	Ensuring fair user data access can compromise data security boundaries, posing organizational security risks (Leslie et al., 2024 )
Governance vs. Inclusiveness	■ Reinforcing	Governance frameworks establish inclusive participation, ensuring representation of various groups in AI decision-making (Bullock et al., 2024 )	Governance mandates diverse stakeholder involvement to ensure inclusiveness in AI policy development (Zowghi & Da Rimini, 2024 )
Governance vs. Privacy	■ Tensioned	Governance mandates can challenge privacy priorities, as regulations may require data access contrary to privacy safeguards (Mittelstadt, 2019 )	Regulatory monitoring demands could infringe on personal privacy by requiring detailed data disclosures for compliance (Solow-Niederman, 2022 )
Inclusiveness vs. Interpretability	■ Reinforcing	Interpretability enriches inclusiveness by ensuring AI systems are understandable, fostering wide accessibility and equitable application (Shams et al., 2023 )	Interpretable AI frameworks enable diverse communities’ meaningful engagement by clarifying system decisions, supporting inclusive practices (Cheong, 2024 )
Inclusiveness vs. Robustness	■ Tensioned	Inclusiveness may compromise robustness by prioritizing accessibility over resilience in challenging scenarios (Leslie, 2019 )	AI built for varied groups might underperform in rigorous environments, highlighting tension (Leslie, 2019 )
Inclusiveness vs. Safety	■ Reinforcing	Inclusiveness motivates safety by ensuring diverse needs are considered, enhancing overall safety standards (Fosch-Villaronga & Poulsen, 2022 )	Inclusive AI health tools accommodate diverse groups, improving overall safety and personalized healthcare (World Health Organization, 2021 )
Inclusiveness vs. Security	■ Tensioned	Inclusiveness in AI can expose it to vulnerabilities, challenging security measures (Fosch-Villaronga & Poulsen, 2022 ; Zowghi & Da Rimini, 2024 )	Inclusive AI systems might prioritize accessibility but compromise security, as noted when addressing diverse infrastructures (Microsoft, 2022 )
Interpretability vs. Privacy	■ Tensioned	Privacy constraints often limit model transparency, complicating interpretability (Cheong, 2024 )	In healthcare, strict privacy laws can impede clear interpretability, affecting decisions on patient data (Wachter & Mittelstadt, 2019 )
Privacy vs. Robustness	■ Tensioned	Achieving high privacy can sometimes challenge robustness by limiting data availability (Hamon et al., 2020 )	Differential privacy techniques may decrease robustness, impacting AI model performance in varied conditions (Hamon et al., 2020 )
Privacy vs. Safety	■ Tensioned	Balancing privacy protection with ensuring safety can cause ethical dilemmas in AI systems (Bullock et al., 2024 )	AI in autonomous vehicles must handle data privacy while addressing safety features (Bullock et al., 2024 )
Privacy vs. Security	■ Reinforcing	Both privacy and security strive for safeguarding sensitive data, aligning objectives (Hu et al., 2021 )	Using encryption methods, AI systems ensure privacy while maintaining security, protecting data integrity (Hu et al., 2021 )
Pillar: Ethical Safeguards vs. Societal Empowerment
Accountability vs. Human Oversight	■ Reinforcing	Accountability necessitates human oversight for ensuring responsible AI operations, requiring active human involvement and supervision (Leslie, 2019 )	AI systems in healthcare employ human oversight for accountable decision-making, preventing potential adverse outcomes (Novelli et al., 2024 )
Accountability vs. Sustainability	■ Reinforcing	Accountability aligns with sustainability by ensuring responsible practices that support ecological integrity and social justice (van Wynsberghe, 2021 )	Implementing accountable AI practices reduces carbon footprint while enhancing brand trust through sustainable operations (van Wynsberghe, 2021 )
Accountability vs. Transparency	■ Reinforcing	Transparency supports accountability by enabling oversight and verification of AI systems’ behavior (Dubber et al., 2020 )	In algorithmic finance, transparency enables detailed audits for accountability, curbing unethical financial practices (Dubber et al., 2020 )
Accountability vs. Trustworthiness	■ Reinforcing	Accountability builds trustworthiness by enhancing transparency and integrity in AI operations (Schmidpeter & Altenburger, 2023 )	AI systems with clear accountability domains are generally more trusted in healthcare settings (Busuioc, 2021 )
Bias Mitigation vs. Human Oversight	■ Reinforcing	Human oversight supports bias mitigation by ensuring continual auditing to detect and address biases (Ferrara, 2024 )	In hiring AI, human oversight helps identify bias in training data biases, enhancing fairness (Ferrara, 2024 )
Bias Mitigation vs. Sustainability	■ Reinforcing	Bias mitigation supports sustainability by fostering fair access to AI benefits, reducing societal imbalances (Rohde et al., 2023 )	Ensuring AI equitable data distribution reduces systemic biases, contributing to sustainable growth (Rohde et al., 2023 )
Bias Mitigation vs. Transparency	■ Reinforcing	Bias mitigation relies on transparency to ensure fair AI systems by revealing discriminatory patterns (Ferrara, 2024 )	Transparent algorithms in recruitment help identify bias in decision processes, ensuring fair practices (Ferrara, 2024 )
Bias Mitigation vs. Trustworthiness	■ Reinforcing	Bias mitigation fosters trustworthiness by addressing discrimination, thereby improving user confidence in AI systems (Ferrara, 2024 )	In lending AI, bias audits enhance algorithm reliability, fostering trust among users and stakeholders (Ferrara, 2024 )
Fairness vs. Human Oversight	■ Reinforcing	Human oversight supports fairness by ensuring AI decisions reflect equitable practices grounded in human judgment (Voeneky et al., 2022 )	For recruitment AI, human oversight calibrates fairness, reviewing bias mitigation strategies before final implementation (Bateni et al., 2022 )
Fairness vs. Sustainability	■ Reinforcing	Fairness supports sustainability by advocating equitable resource distribution, essential for sustainable AI solutions (Schmidpeter & Altenburger, 2023 )	AI systems ensuring fair access to renewable energy results underscore this synergy (van Wynsberghe, 2021 )
Fairness vs. Transparency	■ Reinforcing	Transparency in AI increases fairness by allowing for the identification and correction of biases (Ferrara, 2024 )	Transparent hiring algorithms enable fairness by revealing discriminatory patterns in recruitment practices (Lu et al., 2024 )
Fairness vs. Trustworthiness	■ Reinforcing	Fairness enhances trustworthiness by promoting equal treatment, diminishing bias, thus fostering confidence in AI systems (Cheong, 2024 )	Mortgage AI with fair credit evaluations strengthens trustworthiness, ensuring non-discriminatory decisions for applicants (Dubber et al., 2020 )
Human Oversight vs. Inclusiveness	■ Reinforcing	Human oversight promotes inclusiveness by ensuring diverse perspectives shape AI ethics and implementation (Dubber et al., 2020 )	Human oversight in AI enhances inclusiveness by involving diverse stakeholder consultations during system development (Zowghi & Da Rimini, 2024 )
Human Oversight vs. Privacy	■ Tensioned	Human oversight might collide with privacy, requiring access to sensitive data for supervision (Solove, 2025 )	AI deployment often requires human oversight conflicting with privacy norms to evaluate sensitive data algorithms (Dubber et al., 2020 )
Inclusiveness vs. Sustainability	■ Reinforcing	Inclusive AI development inherently supports sustainable goals by considering diverse needs and reducing inequalities (van Wynsberghe, 2021 )	AI initiatives promoting inclusivity often align with sustainability, as seen in projects that address accessibility in green technologies (van Wynsberghe, 2021 )
Inclusiveness vs. Transparency	■ Reinforcing	Both inclusiveness and transparency promote equitable access and understanding in AI, enhancing collaborative growth (Buijsman, 2024 )	Diverse teams enhance transparency tools in AI systems, ensuring fair representation and increased public understanding (Buijsman, 2024 )
Inclusiveness vs. Trustworthiness	■ Tensioned	Trust-building measures, like rigorous security checks, can marginalize less-privileged stakeholders (Bullock et al., 2024 )	Expensive trust audits in AI systems may exclude smaller organizations from participation (Dubber et al., 2020 )
Privacy vs. Sustainability	■ Tensioned	Privacy demands limit data availability, hindering AI’s potential to achieve sustainability goals (van Wynsberghe, 2021 )	Strict privacy laws restrict data collection necessary for AI to optimize urban energy use (Bullock et al., 2024 )
Privacy vs. Transparency	■ Tensioned	High transparency can inadvertently compromise user privacy (Cheong, 2024 )	Algorithm registries disclose data sources but risk exposing personal data (Buijsman, 2024 )
Privacy vs. Trustworthiness	■ Reinforcing	Privacy measures bolster trustworthiness by safeguarding data against misuse, fostering user confidence (Lu et al., 2024 )	Adopting privacy-centric AI practices enhances trust by ensuring user data isn’t exploited deceptively (Lu et al., 2024 )