{"resourceId":"frontiers-li-ai-coaches-ethics-2026-03-30","versions":[{"version":"external-36f9879e249b6709aac1ab43a1c1dd82fd26e1516babbde3a77af4cb2cd4faea","resource":{"id":"frontiers-li-ai-coaches-ethics-2026-03-30","title":"Frontiers: AI coaches raise privacy, bias, and responsibility gaps needing human oversight","organization":"Shanghai University of Sport; Shanghai University of Finance and Economics; Nanjing University of Finance and Economics","sector":"Sport AI ethics (conceptual; not NCAA-specific trial)","geography":"China (normative ethics analysis with global regulatory references)","publishedAt":"30 March 2026","publicationDate":"2026-03-30","eventDate":null,"sourceName":"Frontiers in Digital Health","sourceLabel":"Peer-reviewed academic research (conceptual/normative)","sourceUrl":"https://www.frontiersin.org/journals/digital-health/articles/10.3389/fdgth.2026.1781352/full","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["governance-procurement","data-security","accessibility-workforce","operating-model"],"finding":"Li, Bian, and Wu (published 2026-03-30) argue AI coaching systems that continuously collect wearable/app athlete data create three primary ethical risks—privacy invasion, data bias, and unclear responsibility after harm—and recommend privacy-by-design (minimization, encryption, consent), bias auditing/diverse datasets, and multi-party responsibility with human coaches retaining final decisions on high-risk guidance. Why now: U.S. college athletics is scaling communications and coaching assistants (e.g., FanWord, Flex) where these controls are often underspecified in announcements.","sledRelevance":"Independent ethical scrutiny for AI coaching relevant to collegiate athlete-support and performance tooling; not a measured NCAA trial, but specifies controls athletics should demand before autonomous training advice.","evidence":"Conceptual/normative analysis with literature synthesis and an illustrative personalized-training AI-coach case; proposes governance measures referencing GDPR/UNESCO/China PIPL rather than new empirical field results.","architectureImplications":"Design AI coaches as human-in-the-loop systems with data minimization, anonymization, and mandatory human approval for return-to-play or high-risk load changes.","governanceImplications":"Assign duties across developers, operators, sports organizations, coaches, and athletes; create ethics review and third-party audit paths before deployment.","securityPrivacyImplications":"Warns that leaks of training/health/location data can enable competitive misuse, commercial reuse, and safety threats; consent and deletion rights are required controls.","caveats":"Not an empirical evaluation of a named NCAA product; China-based authorship with illustrative cases; does not quantify college-athletics incident rates.","streamIds":["college-athletics"],"roles":{"sales":"Athletics leaders buying AI coaching or S&C assistants need ethics and liability clarity, not only personalization demos. Discovery should ask what athlete data is collected, whether outputs can change training without human sign-off, and who is accountable after injury. Bound value is a controlled co-pilot for routine cues—not autonomous coaching. Unsupported claims include bias-free recommendations, guaranteed injury prevention, or that vendor terms alone assign responsibility.","engineering":"Fit is assistive coaching under human authority. Architecture should minimize sensors/data fields, encrypt storage, log recommendations, and require coach acknowledgment for high-risk adaptations. Prerequisites: consent UX, bias evaluation plan, and role-based access. Constraints: conceptual paper without product benchmarks. Security: block secondary advertising use and competitor-accessible exports of athlete weakness data. Proof of value: completed DPIA, bias audit sample, and forced human approval on rehab/load alerts.","delivery":"Delivery work is policy + product configuration: data inventory, consent scripts, coach training, and incident drills. Dependencies include sports medicine and compliance ownership. Ownership: athletics performance/medicine for clinical judgment; vendor for model safety. Skills adoption teaches coaches to override AI and document why. Governance checkpoints: pre-season ethics review, quarterly bias/privacy audits, and clear athlete complaint channels. Acceptance criteria: no autonomous high-risk plan changes; published responsibility matrix. Risks include liability ambiguity and biased plans harming underrepresented athletes."},"roleDetails":{"sales":{"customerProblem":"Departments adopting AI coaches lack clear answers on privacy, fairness, and who is liable when AI-guided training contributes to harm.","stakeholders":["AD / SWA","Sports performance and sports medicine","Compliance / privacy office","Student-athletes and parents"],"discoveryQuestions":["Which data categories feed the AI coach, and for what retention period?","Can the system change load or rehab plans without human approval?","Where is responsibility allocated in the contract after an adverse event?"],"valueHypothesis":"Ethically governed AI coaching can extend personalized cues if humans retain authority and data uses are minimized.","potentialEngagement":"Ethics-and-controls workshop producing a responsibility matrix and minimum privacy requirements before pilot.","unsupportedClaims":["Empirical proof the authors’ framework reduces injuries","Bias-free AI coaching","NCAA-ready compliance certification from this paper alone"]},"engineering":{"fit":"Human-supervised AI coaching with privacy-by-design—not fully autonomous goal-setting systems.","architectureIntegration":"Sensor/app ingest → minimized feature store → recommendation engine → coach acknowledgment gate → athlete device.","prerequisites":["Informed consent and deletion workflows","Bias/fairness evaluation dataset plan","Audit logging of recommendations"],"constraints":["Paper is conceptual, not a product bake-off","Cross-jurisdiction privacy rules differ","Autonomy claims increase liability complexity"],"security":"Encrypt athlete data, prevent advertising reuse, and restrict competitive-sensitive training details to need-to-know roles.","proofOfValue":["DPIA completed before production","Documented human approval on high-risk alerts","Bias audit across sex/region subgroups used in training data"]},"delivery":{"work":"Implement consent, minimization, coach override training, and an incident/responsibility playbook.","dependencies":["Sports medicine sign-off","Vendor contractual liability clauses","Campus privacy office"],"ownership":"Sports performance owns day-to-day use; compliance owns privacy policy; vendor owns model safety patches.","skillsAdoption":"Train coaches to treat AI as draft guidance and to record overrides.","governanceCheckpoints":["Pre-season ethics review","Quarterly privacy/bias audit","Post-incident responsibility drill"],"acceptanceCriteria":["Written ban on autonomous high-risk plan changes","Athlete-accessible data view/delete path","Responsibility matrix signed by AD and vendor"],"risks":["Unclear liability after injury","Biased plans for underrepresented athletes","Location/health data leakage"]}},"retrievedAt":"2026-09-28T03:18:00Z","enrichedAt":"2026-09-28T03:29:00Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Positions AI coaches as potentially expanding access to guidance when human coaches are scarce, while warning deskilling and unfairness if bias and responsibility are ignored.","procurementImplications":"Require privacy-by-design exhibits, bias testing, and contractual responsibility allocation—not only personalization demos—before AI coaching pilots.","operatingModelImplications":"Institutionalizes human-in-the-loop coaching: AI may draft plans, but coaches/clinicians own high-risk decisions.","sourceVerification":{"openedUrl":"https://www.frontiersin.org/journals/digital-health/articles/10.3389/fdgth.2026.1781352/full","referenceExcerpt":"significant ethical concerns, including privacy violations, data biases, and ambiguous responsibility in cases of failure or misuse","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}