The companion control catalogue to The Unconstrained Machine by John M. Waller.
The book makes the case for deliberate limits on AI. This library organizes the controls that support that work: 272 controls organized into 14 families. It gives readers and practitioners a common reference for identifying what needs to be governed, protected, monitored and constrained.
How to use the catalogue
Choose a family below to read its purpose and browse its complete list of control identifiers and titles. The identifiers correspond to Appendix I of the book, making it easy to move between the printed catalogue and this online reference.
The catalogue contains 191 core controls and 81 supplemental controls. These designations organize the library; applicability depends on the system, its use and its risks. Supplemental controls may be essential in a particular environment.
The four control types
The Unconstrained Machine describes four complementary ways to constrain AI. Each addresses a different source of limits on what a system can do and how it is used.
| Control type | Meaning in the book |
|---|---|
| Technical | Hard limits built into the system |
| Governance | Hard external rules, laws and regulations |
| Operational | Soft external norms, policies and processes |
| Contextual | Soft limits from the environment |
Here, hard and soft describe how a constraint operates, rather than how important it is. A policy or process can guide behavior, while a technical mechanism can enforce a boundary. External laws and regulations establish binding obligations, and the surrounding environment can shape behavior and decisions.
The 14 families below organize controls by subject. They are distinct from the four control types: a family can bring together controls of several types, and a control can rely on more than one type in practice. The family descriptions illustrate these relationships; they do not assign a formal type to each individual control.
Browse the 14 control families
- CF-01: Lifecycle Governance (34 controls)
Establish who owns the AI program, how risks are assessed, and what must happen before a system or material change is approved. - CF-02: Fairness, Bias & Rights Impact (13 controls)
Identify who could be affected, define relevant fairness measures, and establish how findings will change deployment decisions. - CF-03: AI Supply Chain & Third-Party Risk (18 controls)
Inventory the components and providers behind each AI service, then document which party is responsible for each relevant safeguard. - CF-04: Auditability & Evidence (19 controls)
Determine what evidence is needed to reconstruct a decision or action, and ensure the records are protected and available to authorized reviewers. - CF-05: Data Governance & Privacy (36 controls)
Trace the data used by each system, identify its owner and sensitivity, and define permitted uses and retention before expanding access. - CF-06: Transparency & Explainability (12 controls)
Identify each audience and the decisions it must make, then provide clear information about AI involvement, limitations and available oversight. - CF-07: Human Oversight & Escalation (25 controls)
Assign intervention authority, define when human review is required, and test whether operators can override or stop consequential activity. - CF-08: Testing & Robustness (21 controls)
Define meaningful acceptance criteria for the use case and test the complete application, including its data paths, integrations and permissions. - CF-09: Monitoring & Drift Detection (12 controls)
Establish a baseline, define actionable thresholds, and ensure that each meaningful alert has an owner and a response path. - CF-10: Recourse & Incident Response (13 controls)
Define what counts as an AI incident, establish reporting and escalation paths, and give affected people an accessible way to challenge consequential outcomes. - CF-11: Objective & Policy Constraints (11 controls)
Write down what the system may do, what it must never do, and which technical mechanisms enforce those limits. - CF-12: Security & Abuse Resistance (34 controls)
Map the application's identities and access paths, minimize privileges, and enforce boundaries around data retrieval and tool execution. - CF-13: Model & Adversarial Security (11 controls)
Identify which model assets and attack paths matter in the deployment, then test prevention and detection against those threats. - CF-14: Agentic & Autonomous System Controls (13 controls)
Define each agent's authority and resources, trace delegated work, and verify that boundaries and termination apply across the full workflow.
About this edition
This launch edition provides the complete control inventory and descriptions of all 14 families. Detailed implementation guidance, verification methods, system-type applicability and escalation rules for individual controls are planned as a separate expansion.
The catalogue accompanies the book and can evolve as the library develops. Catalogue version 1.0 contains 272 controls. Family descriptions were revised on 31 October 2026 to reflect the book’s four control types.