AG-815
Sports and Esports Integrity Governance
EU AI Act
NIST AI RMF
ISO 42001
AGS Sector Governance | Sports, Esports & Athletic Integrity | Version 2.2
Sports and Esports Integrity Governance governs AI agents used for competition integrity — match-fixing and betting-anomaly detection, anti-doping analytics, and athlete performance/biometric analysis — requiring human oversight of any sanctioning decision, consent and protection for athlete biometric data, and bias safeguards in integrity models.
AI increasingly supports integrity functions whose outputs can end careers or impose sanctions. This dimension ensures such agents are accurate, fair, privacy-respecting, and never the sole basis for a sanction, alongside the cross-cutting AGS controls.
In scope: integrity-detection agents (match-fixing, betting anomalies, doping suspicion); athlete biometric/performance data protection and consent; human oversight of sanctioning; fairness/bias and evidence-verification safeguards.
Out of scope: general sports-media/content agents and ticketing/commerce. This dimension governs *AI agents affecting competition integrity and athlete rights*.
An integrity model's output can trigger a doping case, a fixing investigation, or a competitive ban — decisions with profound consequences for individuals. A false positive driven by biased data or unverified signals can wrongly destroy a career; a false negative lets corruption persist. Athlete biometric data is highly sensitive. Governance ensuring human-adjudicated sanctions, consented data use, and bias control is essential to fair, defensible integrity operations.
Test 6.1: Human-Adjudicated Sanction
Test 6.2: Bias Evaluation
Test 6.3: Data Consent & Minimisation
| Score | Criteria |
|---|---|
| 0 | AI integrity outputs drive sanctions automatically; athlete data ungoverned |
| 1 | Human review exists but no bias evaluation / weak data governance |
| 2 | Human-adjudicated sanctions, bias-evaluated models, evidence corroboration, contest routes, logs |
| 3 | Error-rate monitoring, detector-manipulation resistance, due-process access, minimised secured biometrics |
Scenario A — Wrongful Ban: A doping-suspicion model flags an athlete on a spurious correlation; an automated process imposes a provisional ban. Human adjudication with corroborated evidence would have prevented the wrongful sanction.
Scenario B — Biased Suspicion: A fixing model disproportionately flags athletes from one region due to skewed training data. Bias evaluation would have surfaced and mitigated the disparity.
Scenario C — Biometric Leak: Athlete biometric data collected for performance analysis is retained indefinitely and breached. Minimisation, security, and retention limits would have reduced the harm.
| Requirement | EU AI Act | NIST AI RMF | ISO 42001 |
|---|---|---|---|
| R1: Human-adjudicated sanctions | Art. 14 — Human oversight | MAP 3.5 — Human oversight | A.9 — Use of AI systems |
| R2: Biometric data consent/minimisation | Art. 10 — Data governance | MEASURE 2.10 — Privacy risk | A.7 — Data for AI systems |
| R3: Bias evaluation | Art. 10 — Bias examination | MEASURE 2.11 — Fairness and bias | A.5 — Impact assessment |
| R4: Evidence corroboration | Art. 15 — Accuracy | MEASURE 2.6 — Safety/validity | Clause 8.3 — Verification |
| R5: Contest route / due process | Art. 14 — Human oversight | GOVERN 5.1 — External feedback | A.8 — Information for interested parties |
| R6: Tamper-evident adjudication log | Art. 12 — Record-keeping | MEASURE 2.4 — Production monitoring | Clause 8.1 — Operational control |
| R7: Detector-manipulation resistance | Art. 15 — Robustness | MEASURE 2.7 — Security and resilience | Clause 8.3 — Verification |
| R8: Error-rate monitoring | Art. 15 — Accuracy | MEASURE 2.6 — Safety/validity | Clause 9.1 — Monitoring and measurement |
Article 14 (human oversight) ensures AI integrity outputs do not auto-sanction; Article 10 (data governance, incl. bias examination) governs sensitive athlete data and model bias. AG-815 applies both to competition-integrity agents.
MEASURE 2.11 (fairness and bias) and MAP 3.5 (human oversight) frame fair, human-adjudicated integrity decisions.
Clause 6.1 (actions to address risks) and Annex A.5 (impact assessment) require assessing and mitigating the individual-rights impacts of integrity AI.