Ethics dumping in artificial intelligence

“Most ethical AI principles articulate high-level injunctions like transparency, fairness, accountability, and respect for human rights that few would disagree with (Jobin et al., 2019).”

“…the opacity and inscrutability of most AI systems leave users with little insight into the rationale or assumptions driving system behaviors, and leave them with little means of addressing the normative considerations baked into the algorithms (Burrell, 2016; Akhai, 2023; Stafie et al., 2023).”

“In the wake of rising concerns over AI’s societal impacts, a flourishing ecosystem of scholarship and initiatives has emerged to develop ethical frameworks and guidelines for responsible AI development and use (Hagendorff, 2020).”

“As the discourse matures, increasing attention is devoted to the challenges of translating ethical principles into practice and developing robust governance mechanisms (Mittelstadt, 2019; Ryan and Stahl, 2020).”

“However, ethical issues like eroding human decision autonomy, entrenching biases from historical data, and disrupting professional role boundaries often remain underexplored during acquisition and piloting phases (Char et al., 2018).”

“Well-intentioned efforts to leverage transformative AI capabilities can prompt adoption of systems whose embedded normative logics and accountability implications were not robustly evaluated from an ethical lens (Taddeo and Floridi, 2018).”

“…norms within AI systems’ architecture (designed in a specific cultural/institutional milieu), users adopt this “black boxed” normative baggage upon deployment even if antithetical to local values or accountability customs (Ananny and Crawford, 2018; Crawford, 2021; Bélisle-Pipon et al., 2021).”

“Finally, the very distance between developers and where impacts manifest means developers are not the ones who will have to manage the impacts of their technology (Crawford and Calo, 2016).”

“As highly opaque and autonomous systems, developers often have disproportionate influence in shaping AI’s ethical dimension by encoding inferences, assumptions, and values into algorithms and training datasets (Rahwan, 2018; ; McDermid et al., 2021).”

“AI’s undeniable power has sparked discussions in domains such as ELSI (ethical, legal, and social implications), STS (Science and Technology Studies), and applied ethics (Dubber et al., 2020; Hagendorff, 2020; Horgan et al., 2020; Bélisle-Pipon et al., 2021; Slota et al., 2021).”

“Key concerns include responsibly embedding ethics into design and deployment pipelines (Floridi et al., 2020), evaluating normative impacts across contexts (Eitel-Porter, 2020), and establishing accountability structures (Russell et al., 2015).”

“What fuels AI’s transformative promise and “hype” is precisely its potential to significantly reshape practices, beliefs, and decision-making processes (Matheny et al., 2019; Char et al., 2020).”

“…section outlines four key trends illustrating how practices from different actors across the AI ecosystem can contribute to ethics dumping. by numerous organizations aiming to shape the normative discourse and establish ethical guardrails (Whittlestone et al., 2019; Keller and Drake, 2020).”

“Some suggest that ethics-by-design can resolve these tensions by hardwiring ethical guidelines into algorithms from inception (Dignum, 2018).”

“…predictive and diagnostic tools, clinical decision support systems (CDSS), AI-generated in-basket responses, or patient monitoring tools tout promises of increased efficiency and optimized resource allocation (Chen et al., 2021; Elhaddad and Hamam, 2024; Garcia et al., 2024; Khosravi et al., 2024).”

“…predictive and diagnostic tools, clinical decision support systems (CDSS), AI-generated in-basket responses, or patient monitoring tools tout promises of increased efficiency and optimized resource allocation (Chen et al., 2021; Elhaddad and Hamam, 2024; Garcia et al., 2024; Khosravi et al., 2024).”

“Key concerns include responsibly embedding ethics into design and deployment pipelines (Floridi et al., 2020), evaluating normative impacts across contexts (Eitel-Porter, 2020), and establishing accountability structures (Russell et al., 2015).”

“Mitigating this requires a pull rather than push approach to AI deployment, where systems are co-developed or iteratively customized to meet verified needs of host environments and align with in-situ value systems (Shilton, 2018).”

“This necessitates the establishment of independent oversight bodies with the authority to enforce sanctions and conduct investigations into breaches of ethics (Schmitt, 2022).”

“However, despite this flurry of activity articulating principles and guidelines, concerns remain over how well AI ethics translates into practice (Ho, 2020).”

“Trend 1: ethics dumping by AI developers Like all technologies, AI systems have profound impacts on users and their environments that go beyond just technical functionality (Alami et al., 2020; ÓhÉigeartaigh et al., 2020).”

“Second, prescriptive top-down design and ruling force a “one-size-fits-all” paradigm on diverse contexts, overriding local values and accountability customs (Cinnamon, 2019).”

“AI’s undeniable power has sparked discussions in domains such as ELSI (ethical, legal, and social implications), STS (Science and Technology Studies), and applied ethics (Dubber et al., 2020; Hagendorff, 2020; Horgan et al., 2020; Bélisle-Pipon et al., 2021; Slota et al., 2021).”

“However, this initial wave prioritized articulating high-level principles over examining practical challenges in norm design and implementation (Lauer, 2020).”

“…the outset (and not be an afterthought) of the development and deployment of an AI solution, as well as adopting an inclusive, open and transparent approach, especially when it affects a very large number of people or is a public/ governmental solution (Dobbe et al., 2021; Couture et al.,2023).”

“Finally, such governance frameworks frequently concentrate liability risk and regulatory compliance responsibilities on frontline users rather than on developers and institutional procurers of AI systems (Duffourc et al., 2023).”

“Local norms, constraints, and stakeholder values risk being overridden by one-size-fits-all frameworks (Chaudhary, 2020).”

“At its core, ethics dumping emerges from skewed power dynamics, where certain downstream populations or environments are unable to negotiate or uphold ethical practices aligned with their values and contexts (Schroeder et al., 2019b; Samuel and Derrick, 2020).”

“Responsible AI development cannot presume to dictate conditions of use but must empower recipients with meaningful refusal rights and agenda-setting influence over ethical baselines (Tinnirello, 2022).”

“This is especially true for complex and impactful technologies like artificial intelligence (AI) (Elliott, 2019; Alami et al., 2020).”

“…we look to the concept of “ethics dumping” in research ethics, which highlights how the export of ethical guidelines and governance from powerful actors to less-privileged contexts can impose disproportionate costs and burdens if local conditions are not accounted for (Schroeder et al. 2019a).”

“” This can occur when limited capacity and expertise hinder effective governance, leading to insufficient protections for those in need (Schroeder et al., 2019b).”

“…from key stakeholder groups—including subcontractors, site workers, and regulators—during AI design and implementation leads to poor alignment with real project conditions and reinforces structural inequalities, creating disadvantages for underrepresented groups (Bélisle-Pipon and Victor 2024).”

“This trend aligns with concerns over an impending “AI regulation winter” (Bélisle-Pipon, 2024) scenario in which regulatory mechanisms are intentionally weakened to serve elite interests, further entrenching their dominance.”

“• The second is the knowledge commons, where the appropriation of human-authored content for AI training can generate epistemic asymmetries, displace creative labor, and reduce plural contributions [5, 18].”

“Developers, clinicians, patients, policymakers, and civil society actors are presented not as passive recipients of ethical guidance, but as active agents responsible for aligning technological deployment with public values (Vayena et al., 2018; Bélisle-Pipon and Victor, 2024).”

“While research discussing the ethics of AI usage tends to be conceptual or normative, it does not always empirically map how the theme evolves in scientific literature (Bélisle-Pipon & Victor, 2024 ; Cano, 2025 ; Giarmoleo et al., 2024 ; Saxena, 2025 ).”

“This transformation not only brings improvements in functionality and efficiency but also raises ethical issues and uncertainties (Askin et al., 2023; Bélisle-Pipon & Victor, 2024).”

“Avoiding such early engagement risks AI ethics dumping, deferring ethical accountability and governance to later stages, thereby offload - ing the burden of addressing core concerns onto downstream actors ill-equipped to rectify foundational oversights [ 70].”

“As noted elsewhere, trustworthy AI—particularly within healthcare— necessitates coordinated efforts among developers, policymakers, and (clinical) institutions to uphold ethical standards, transparency, and accountability (Bélisle-Pipon, 2024).”

“Among these are fears of excessive dependency, reduced cognitive effort, uncritical acceptance of responses, ethical dumping, risk of inaccuracies or errors, and lack of pedagogical mediation in the use of the tool [4], [6], [7], [8].”

“Similar dynamics have been described in discussions of “ethics dumping” in AI research, where ethical responsibilities are shifted to environments with weaker oversight (Bélisle-Pipon and Victor 2024).”

“In addition, issues such as writing agency, academic responsibility, and academic integrity are often discussed normatively (Bélisle-Pipon & Victor, 2024; Podgornaya et al., 2025; Saxena, 2025; Wahyuni et al., 2026), without delving into how students negotiate the boundaries of AI use in their…”

“This “ethics washing” can legitimize harm ‑ ful practices, delay meaningful regulation, and entrench power imbalances [7, 9, 14–16 ].”

“Ultimately, fostering a culture of openness and responsibility concerning AI technologies is crucial for their ethical and sustainable integration into society [23].”

“It also entails avoiding ethics dumping and instead supporting a culture of care within healthcare institutions (Bélisle-Pipon and Victor 2024); one that recognizes compassion not as an optional add-on, but as a foundational moral gounding.”