Cinderpoint AI

AI Concepts, Programs, and Tools

Concepts define the problems. Programs structure the science. The tools put the work into practice. All published research by Waydell D. Carvalho.

Concepts
Concept
Strategic Silence Problem
Why advanced AI may systematically underdisclose capabilities during evaluation.
Concept
AI Student Test
A learning-centered framework for evaluating generative AI use in higher education.
Concept
Copyright-Piracy Confusion
How conflated terms collapse the policy debate over AI training data.
Concept
Wizard of AI Curtain Test
A diagnostic for separating real model behavior from staged demonstrations.
Concept
Runtime Governance Gap
The accountability failure created when AI systems evolve after deployment.
Outpaced by AI  ·  13 Failure Modes

The thirteen structural ways AI deployments fail inside competent organizations. Each is a canonical concept page drawn from Outpaced by AI: 13 Ways Organizations Risk Deployment and Governance Failure by Waydell D. Carvalho. About the book ›

Failure Mode 1
Illusion of Validity
Passing every test is mistaken for being safe.
Failure Mode 2
Illusion of Control
Owning a system on paper is mistaken for controlling what it decides.
Failure Mode 3
Ceremonial Compliance
Oversight that runs and records, but cannot stop anything.
Failure Mode 4
The Pacing Problem
The system reaches full scale before oversight can constrain it.
Failure Mode 5
Alert Fatigue
The one alert that matters is lost in the thousands that do not.
Failure Mode 6
Normalization of Deviance
Operating outside the rules becomes normal because nothing has failed yet.
Failure Mode 7
Post-hoc Rationalization
The explanation is built after the decision, for whoever asks.
Failure Mode 8
Surrogation
The proxy metric becomes the goal, and the real goal is lost.
Failure Mode 9
The Problem of Many Hands
Everyone owns a piece; no one owns the whole.
Failure Mode 10
Coordination Neglect
The finding arrives accurate, and too late to matter.
Failure Mode 11
Threat Rigidity
Evidence of failure triggers doubling down, not changing course.
Failure Mode 12
Moral Licensing
A public ethics brand becomes cover for the decisions it should check.
Failure Mode 13
Cascading Failures
Forgotten state meets a new change and spreads faster than anyone can react.
Programs
Program
SAIERA
Thermodynamic limits of advanced intelligence. Framework, hypothesis, and applied science program.
All Published Papers
SSRN The Runtime Governance Gap: Governing Self Modifying Artificial Intelligence Through Adaptive Oversight Apr 2026 SSRN Thermodynamic Limits of Advanced Intelligence: The SAIERA Framework Apr 2026 Zenodo The Strategic Silence Problem: Why Advanced AI May Hide Its Capabilities Mar 2026 Zenodo Runtime Governance and Adaptive AI Systems: A Framework for Continuous Oversight Feb 2026 SSRN SAFEARC: A Seven-Pillar Operational Architecture for Socio-Technical AI Governance Jan 2026 SSRN Governing Self-Modifying AI: A Federal Framework for Runtime Safety Aug 2025 Zenodo The AI Student Test: A Learning-Centered Framework for Evaluating Generative AI Use in Higher Education Jan 2026
SSRN Author Page ↗ Zenodo ↗
Recent AI Articles

Latest articles tagged AI. See all articles ›