.png)
Trustworthy AI Starts Before Release
An AI model's risks are largely decided while it is being built: which data it learns from, how it is tested and what limits are placed on its outputs. Problems found after release are harder and more expensive to fix, and AI laws increasingly expect organizations to show this work before a model reaches users.
SCF CORE AI Model Deployment brings together the SCF controls for building and releasing AI responsibly. The SCF maps these controls to AI-specific requirements such as the EU AI Act, the NIST AI Risk Management Framework (AI RMF) and ISO/IEC 42001, so one set of practices can support several obligations.
SCF CORE AI Model Deployment is part of the CORE initiative, which the SCF Council created to turn the comprehensive SCF catalog into focused, actionable baselines. Its controls are derived from the Common Controls Framework (CCF) and calibrated to the work of building and releasing AI models, giving developers a defined starting point they can tailor to their own models and use cases.
Defined Purpose
Each model has a documented business case, intended use and impact assessment.
Tested Before Release
Models are tested, evaluated, validated and verified against trustworthiness, safety, security and fairness criteria.
Accountable Operation
Deployed models are monitored, documented and supported by clear human oversight.
Volunteer-Driven & Creative Commons Licensed
SCF CORE AI Model Deployment is developed by volunteer cybersecurity practitioners and released under a Creative Commons license. There is no cost to access or implement it.
Expectations Across The AI Lifecycle
Examples of what SCF CORE AI Model Deployment expects include:
Purpose & Business Case
Defined goals, intended scope, expected benefits and costs, and assigned responsibilities
Impact & Risk
Impact assessments, AI threat modeling, likelihood and impact analysis, and data protection impact assessments
Data
Identification, integrity and lineage of training data sources, data quality and de-identification
Development
Documented development practices, requirements and implementation details
Testing
Test, evaluation, validation and verification (TEVV) covering trustworthiness, safety, security, privacy, fairness and bias
Output Safeguards
Output marking, output filtering, human moderation and protection against intellectual property infringement
Resilience
Protection against model pollution, cascading hallucinations and resource exhaustion
Conformity
Conformity assessment, declarations of conformity, registration of AI systems and technical documentation
Transparency
Notification to users and employees when AI is in use
Oversight & Monitoring
Human oversight, post-deployment monitoring, incident reporting and stakeholder feedback
AI Agents
Governance over AI agents, including permissions, sandboxing and kill switch controls
The complete control set is in the SCF download and is updated with each SCF release.
Organizations Building & Releasing AI
SCF CORE AI Model Deployment is designed for organizations that create AI models or place them in products and services used by others.
AI Developers
Organizations that build or fine-tune AI models for their own products or for customers.
AI-Powered Product Companies
Organizations that embed AI models in software, devices or services they sell.
Organizations Subject To AI Laws
Organizations that need to demonstrate conformity with AI-specific laws and standards.
%20(white).png)