Standard for the Duty of Care in Artificial Intelligence
The ADS-2026 standard defines the mandatory Duty of Care requirements for entities developing, deploying, or operating Artificial Intelligence systems with significant societal impact.
All AI systems must be optimized for the preservation and enhancement of human physical, psychological, and social wellbeing.
Systems must undergo a mandatory "Human Impact Assessment" (HIA) prior to deployment.
Optimization functions must explicitly exclude metrics that incentivize psychological harm or addiction.
Transparency is the foundation of accountability. Entities must maintain verifiable records of system behavior and data provenance.
Entities must maintain immutable "Inference Logs" for all high-stakes autonomous decisions.
Training data lineage must be documented and accessible to accredited third-party auditors.
Model limitations and known failure modes must be disclosed via standardized "ADS Model Cards".
AI systems must not be used to manipulate public discourse or amplify systemic biases.
Algorithmic preference for specific political or commercial ideologies is strictly prohibited.
Continuous bias monitoring and mitigation protocols must be active during the entire lifecycle.
Robust fail-safes and circuit breakers must be integrated into the core architecture of critical systems.
Mandatory "Hardware Kill-Switch" for AI systems controlling physical infrastructure.
Real-time "Out-of-Distribution" (OOD) detection must trigger immediate human-in-the-loop intervention.
Click on any Verifiable Requirement (VR) to view community feedback or submit your own technical amendment.