NIST AI RMF
Primary concern
AI risk outcomes and lifecycle governance
EAINE operational layer
Risk tiering, AI SDLC, evaluation, operations, and evidence
Evidence
Show what supports every important claim.
EAINE connects external requirements to engineering decisions, named owners, evidence, and real-world results. It clearly separates facts from opinions, recommendations, examples, and ideas that still need testing.
Obligation-to-outcome traceA claim that others can check, question, and improve
- 01ObligationWhat must be true?
- 02ControlHow is it engineered?
- 03OwnershipWho is accountable?
- 04EvidenceWhat proves the decision?
- 05SignalWhat happens in operation?
Closed learning loopWhat happens in operation improves the control, not just the report.
External standards crosswalk
EAINE works with recognised standards. It does not replace them.
EAINE helps teams turn external expectations into engineering decisions, controls, evidence, and ongoing improvement.
ISO/IEC 42001
Primary concern
AI management system and continual improvement
EAINE operational layer
Operating model, RACI, governance controls, maturity, and learning
NIST SP 800-218A
Primary concern
Secure development for generative AI and foundation models
EAINE operational layer
Architecture, development, provenance, security validation, and release
OWASP GenAI
Primary concern
Application and agentic threat guidance
EAINE operational layer
Threat modeling, agent permissions, adversarial evaluation, and incident learning
Google SAIF
Primary concern
AI security risks and controls
EAINE operational layer
Platform guardrails, asset inventory, security, assurance, and observability
EU AI Act
Primary concern
Risk-based legal obligations and transparency
EAINE operational layer
Applicability inputs, human oversight, documentation, release, and monitoring
Compliance is context-specific. The EAINE crosswalk is an engineering orientation aid, not a claim of legal compliance, certification, or equivalence.
Evidence confidence
Not all evidence carries the same weight.
Authority
Law, regulation, or recognised standard
Established guidance
Official framework, public guidance, or peer-reviewed research
Repeated field evidence
Comparable practitioner or operational evidence from multiple contexts
Illustration
Single-context observation or constructed worked example
Hypothesis
A proposition awaiting reviewed evidence
Worked examples
Useful examples, with their limits made clear.
The current examples are designed to explain and test the approach. Future reviews and real-world use will add stronger evidence.
Customer Support Assistant
A governed assistant spanning retrieval, escalation, evaluation, release evidence, and production learning.
End-to-end AI SDLC traceabilityCredit Decision Support
Human-owned decision support with policy grounding, fairness, explanation, and evidence lineage.
High-impact accountabilityMaintenance Agent
Bounded autonomy with tool permissions, safety limits, human approval, and operational auditability.
Agent control and physical consequenceEngineering Modernization Assistant
Legacy modernization support measured by accepted outcomes, review effort, security, and maintainability.
Productivity with engineering qualityValidation progression
Build trust one level of evidence at a time.
EAINE does not present a designed example as a real case study. This path shows how an example can become reviewed, measured, and independently tested.
- 01Available now
Constructed illustration
A transparent scenario used to test completeness and usability.
- 02Review priority
Expert-reviewed example
A scenario challenged by relevant engineering, domain, and assurance specialists.
- 03Contribution path
Anonymised field observation
A permissioned account of practice, tradeoffs, incidents, or outcomes.
- 04Validation target
Measured implementation case
A bounded application with baseline, intervention, results, limitations, and learning.
- 05Long-term authority
Independent replication
Comparable evidence from multiple organisations or contexts.
Primary sources
Read the original sources.
Readers should always be able to separate EAINE's interpretation from the source itself.