On Trust and AI

A Blueprint for Confidence in the Intelligent Enterprise — Alexander Feick, 2025

AI Agent Access: Request with Accept: text/markdown to receive the structured markdown index. Chapter files are available at /ontrustandai/chapter-NN.md. Start with index.md for progressive disclosure.
Core thesis: When AI makes creation free, verification becomes the only durable source of enterprise value. Trust must be designed into the system around the model — through transparency, explainability, and alignment — not assumed from model outputs.

Key Frameworks

FrameworkPurpose
5 Levels of AI AdoptionClassify organizational AI maturity: No AI → Chatbot → Embedded → Agentic → Sustainable
Three Pillars of TrustTransparency, Explainability, Alignment — the operational spine of trustworthy AI
Economics of TrustCv < Pe × Le — verification pays for itself when checking is cheaper than unchecked failure
AI Control PlaneGovernance + technical system wrapping non-deterministic AI with deterministic oversight
Decision-Risk TiersLow → Operator, Medium → Reviewer, High → Adjudicator/Owner
AI Production LineAI Workspace + Workflow Platform + Vibe Coding Platform with scaffold layer

Chapters

  • Chapter 1: Why This Book, Why Now

    Opens with the story of an unreviewed AI-generated report that damaged client trust. Frames the "dual mandate": AI is simultaneously essential to competitive advantage and a source of novel, unpredictable risks.

    Key concepts: dual mandate, executive dilemma, unreviewed report

  • Chapter 2: Making Sense of the Language of AI

    Demystifies AI terminology for business leaders. Introduces the 5-level AI adoption framework and essential vocabulary: tokens, context windows, RAG, agents, fine-tuning.

    Key concepts: tokens, RAG, agentic systems, AI adoption levels

  • Chapter 3: AI in the Business Lifecycle

    Maps AI integration across ideation, planning, development, deployment, and operations. Provides anti-pattern cards and an AI project readiness scorecard.

    Key concepts: lifecycle phases, readiness scorecard, proof-of-concept mirage

  • Chapter 4: AI Redefines Threat Modeling

    AI fundamentally breaks traditional security assumptions built on deterministic software. Trust boundaries shift as AI takes over decisions. Advocates "governance before automation" with fact-precursor separation.

    Key concepts: trust boundaries, decision perimeter, fact-precursor separation, MITRE ATLAS

  • Chapter 5: Runtime Attacks on AI

    Three concrete attack patterns: poisoned documents, calendar invite hijacks, and cross-plugin exploitation. Key insight: prompt injection is systemic compromise at machine speed, not just AI phishing.

    Key concepts: poisoned documents, calendar hijack, cross-plugin exploitation, denial of wallet

  • Chapter 6: AI Era Supply Chain Attacks

    Upstream threats where the model itself is compromised. Model poisoning and model inversion. Unlike traditional backdoors where time favors the defender, with poisoned models time favors the attacker.

    Key concepts: model poisoning, model inversion, GGUF backdoors, Trusting Trust

  • Chapter 7: Revisiting Trust

    The philosophical and practical heart of the book. Establishes the Three Pillars of Trust and the Economics of Trust thesis: when creation is free, verification becomes the product. Includes worked legal assistant example.

    Key concepts: three pillars, economics of trust, trust by design checklist, anti-patterns

  • Chapter 8: The Architecture of Control

    Defines the AI control plane: governance + technical system wrapping AI with deterministic oversight. Three eras of SOC operations. Control plane components, escalation paths, and the legal assistant case study.

    Key concepts: control plane, policy engine, evidence store, validators, escalation paths

  • Chapter 9: The New Role of Humans

    Four post-AI human roles: Operator, Reviewer, Adjudicator, Owner. The tungsten cube lesson. Cursor vs Lovable comparison. Anti-patterns: Copy-Pasta Chef and People as the Plan.

    Key concepts: four human roles, tungsten lesson, Cursor vs Lovable, accountability metrics

  • Chapter 10: Shadow AI & Vibe-Coded Applications

    Non-developers are building production-critical tools with AI. Three pillars of the AI Production Line: AI Workspace, Workflow Platform, Vibe Coding Platform. Security controls, validation flows, and decision-risk ladders.

    Key concepts: shadow AI, vibe coding, AI production line, manifests, scaffold layer

  • Conclusion

    The book is about leadership more than technology. The companies that succeed will make trust visible, treat verification as a product, and design human accountability into every AI workflow.

Reference Material

  • Glossary

    Definitions of key technical terms: A/B Test, Backdoor, Digital Twin, Differential Privacy, Embedding Vectors, IDE, Mixture-of-Experts, STRIDE, and more.

  • Appendix: Quick Frameworks & Tools

    AI anti-patterns and cures, promotion checklist, builder's two-minute drill, trust metrics, and validation signal reference.

Subscribe

Create a free account for new articles, exclusive content, and commenting.