DARTS - five foundational pillars
D - Data Governance
Clear data provenance and controls for the security, privacy and confidentiality of the data your AI touches. Most AI failures trace back here.
A - Accountability
Clear roles across the AI lifecycle = who owns it, who intervenes, who answers when it goes wrong. "The AI did it" is not a defence.
R - Risk Management
R - Risk Management
AI-specific risk and impact assessments: bias, model drift, hallucination, emergent agent behaviour with mitigations matched to them.
T - Transparency
T - Transparency
R - Risk Management
Clear explanation of how and where you use AI, with testing and monitoring to back it up. Increasingly the law, too.
S - Stakeholders
T - Transparency
S - Stakeholders
Mechanisms for the people affected by your AI to give feedback and raise concerns. Governance no one can question isn't governance.