The 31 infographics and 41 interactive diagrams of the Body of Knowledge, grouped by where they appear. Each infographic has its own page with a text alternative, SVG and PNG downloads, embed snippets and a citation.
| Part | Chapter | Infographic | Data visualisation | Poster | Interactive diagram | Total |
|---|---|---|---|---|---|---|
| The discipline | 00 Preface | 0 | 0 | 0 | 0 | 0 |
| The discipline | 01 Definition | 2 | 0 | 0 | 1 | 3 |
| The discipline | 02 Why Now | 1 | 0 | 0 | 1 | 2 |
| The discipline | 03 Values & Principles | 1 | 0 | 0 | 0 | 1 |
| The discipline | 04 The Stack | 3 | 0 | 0 | 2 | 5 |
| The discipline | 05 Patterns | 1 | 0 | 0 | 33 | 34 |
| The discipline | 06 The Role | 0 | 0 | 0 | 1 | 1 |
| The discipline | 07 Maturity Model | 1 | 0 | 0 | 1 | 2 |
| Reference | 08 Regulatory Map | 2 | 0 | 0 | 1 | 3 |
| Reference | 09 Glossary | 0 | 0 | 0 | 0 | 0 |
| Reference | 10 Reading List | 0 | 0 | 0 | 0 | 0 |
| Foundations | 11 AI Defined | 0 | 0 | 0 | 0 | 0 |
| Foundations | 12 Governance Program | 2 | 0 | 0 | 0 | 2 |
| Foundations | 13 Risk Management | 4 | 0 | 0 | 0 | 4 |
| The lifecycle | 14 Development | 1 | 0 | 0 | 1 | 2 |
| The lifecycle | 15 Deployment | 0 | 0 | 1 | 0 | 1 |
| The lifecycle | 16 Fairness & XAI | 1 | 0 | 0 | 0 | 1 |
| The lifecycle | 17 Incidents | 1 | 0 | 0 | 0 | 1 |
| The lifecycle | 23 AI Agents | 1 | 0 | 0 | 0 | 1 |
| Law and standards | 18 EU AI Act | 0 | 0 | 3 | 0 | 3 |
| Law and standards | 19 Privacy & AI | 0 | 0 | 0 | 0 | 0 |
| Law and standards | 20 Existing Law | 0 | 0 | 0 | 0 | 0 |
| Law and standards | 21 AI Laws Worldwide | 0 | 1 | 0 | 0 | 1 |
| Law and standards | 22 Principles & Standards | 1 | 0 | 0 | 0 | 1 |
Figures that index the whole discipline rather than one chapter.
The whole discipline on one canvas: a central node and eight branches (foundations, values, the stack, patterns, the role, obligations, maturity and the learning path), every node a link. Follow the branch you are weakest in, or read the same map as a list below. Drawn from chapters 01–08.
The Spanish edition of the EU AI Act timeline poster: the same seven lanes and dates, as of 2026-09-24, with its text in Spanish and the official Spanish terms of the Act. Share it with Spanish-speaking teams; the English poster sits in chapters 18 and 08. Drawn from chapters 18 and 08.
The Spanish edition of the EU AI Act operator roles poster: the same questions, roles, duties, evidence and Article 25 loop, with its text in Spanish and the official Spanish role names of the Act. Share it with Spanish-speaking teams; the English poster sits in chapter 18. Drawn from chapter 18.
The Spanish edition of the EU AI Act risk ladder poster: the same rungs, high-risk routes, Article 6(3) filter, GPAI track and side notes, as of 2026-09-24, with its text in Spanish. Share it with Spanish-speaking teams; the English poster sits in chapter 18. Drawn from chapters 18 and 21.
The three questions every AI governance function must answer from live systems, and the stack layers that answer each. If you cannot answer one of them for a system today, that is your first task. Drawn from chapter 01.
The five nested objects the discipline governs, from the model out to the organisation, with the controls the chapter names for each. Check that every object in your estate has its control, because governing one object leaves the others open. Drawn from chapter 01.
Where AI governance engineering sits among the adjacent roles (analyst, platform and assurance), turning governance intent into running controls and evidence. Generated from the Body of Knowledge.
The demand for AI governance and the skills employers ask for, using only the figures chapter 02 cites: the IAPP 2025 report and an analysis of 1,997 US postings updated in August 2026. Read the skills as the brief for a hiring or learning plan: build-and-run skills, not review skills. Drawn from chapter 02.
The obligations that are live now and the dates the Omnibus sets for what follows (new bans and marking, the deferred high-risk deadlines, legacy public-authority systems), as the chapter states them as of 2026-09-24. Pick the next stage and map its obligations to artefacts in chapter 08. Generated from the Body of Knowledge.
The eight values, stated as affirmations, and the six principles, stated as commitments, verbatim from the Thesis. Quote them, do not reword them, when you write your own policy. Drawn from chapter 03.
The five ordered steps that stand the whole stack up thinly, one per layer, for a team of one. Do step one this week. Drawn from chapter 04.
Human oversight as an engineered control: classify each action by consequence, put a designed checkpoint only where the stakes justify the latency, and keep the decision and the oversight metrics as evidence. Decide which actions need a person by consequence, then watch approval rate, time-to-decide and override rate. Drawn from chapter 04.
What happens to the stack when you buy the model instead of training it: boundary evals and the controllable perimeter shrink, while inventory and collected supplier evidence grow behind a due-diligence gate. Budget your controls for bounding the system and evidencing the supplier, not only for testing it. Drawn from chapter 04.
The reference toolchain mapped to the five stack layers, from policy engines and the agent registry up to guardrails, observability and the assurance store. Generated from the Body of Knowledge.
The five layers of the stack in their canonical order, each holding the artefacts it produces, with evidence flowing up into assurance. Read the chapter layer by layer, then build in the order it gives for a team of one. Generated from the Body of Knowledge.
Every pattern in the catalogue, placed on the five stack layers it lives in, as links. Start from the layer you are weakest in and open its patterns. Drawn from chapter 05.
How a model or prompt change moves through an eval gate in CI: capability and adversarial evals against a versioned suite, a threshold that ships the release or blocks it and files the result against the registry. Generated from the Body of Knowledge.
The runtime incident pipeline and kill switch: a signal trips a circuit breaker that revokes one agent’s scope without breaking the fleet, while triage opens a reportable incident on the statutory clock. Generated from the Body of Knowledge.
How the deploy pipeline registers each agent, issues it a scoped workload identity, and files signed evidence an auditor can attribute, with a policy gate that denies the unregistered. Generated from the Body of Knowledge.
A governance rule travels as a versioned policy card that a policy engine evaluates at one gate, recording an allow-or-deny verdict against every change. Start by writing one card for one obligation and wiring it to that gate. Generated from the Body of Knowledge.
The build step emits an AI bill of materials that lands on the registry entry and feeds the vulnerability-matching and documentation consumers. If your build does not emit it, you cannot answer what is running. Generated from the Body of Knowledge.
The model card is generated from real training and eval outputs, checked against a schema at one gate, and attached to the release as control evidence. A hand-written card is documentation; a generated, validated one is a control. Generated from the Body of Knowledge.
Runtime signals are checked against controls continuously, and every check writes an evidence record an auditor can read from the assurance store. Continuous means the evidence is produced by the data path, not a quarterly exercise. Generated from the Body of Knowledge.
The fundamental-rights impact assessment is captured as data at intake, assessed and mitigated, then approved at a human gate and stored machine-readable beside the DPIA. Run it before deploying a high-risk system and keep the record where the auditor looks. Generated from the Body of Knowledge.
One internal control set is the hub that external frameworks map onto, with evidence attached once to the hub and reused for all of them. The crosswalk is an index, not the end state; the hub is what you maintain. Generated from the Body of Knowledge.
Controls, assessments and evidence become OSCAL documents that a validator checks and assurance consumers read without anyone re-typing them. Choose the format your assessor can ingest and generate it from data you already hold. Generated from the Body of Knowledge.
A versioned adversarial suite built from a threat taxonomy runs in CI or on a schedule, and every finding is fixed or accepted on the record and filed as evidence, with fixes feeding back into the suite. Treat red-team findings like test failures, with an owner and a deadline. Generated from the Body of Knowledge.
Input and output guardrails on the model or agent path enforce the same policy card that CI evaluated and emit a decision event on every call to the assurance store and, on a breach, to the circuit breaker. A guardrail without events is a filter, not a control. Generated from the Body of Knowledge.
A breaker trips on a signal and revokes one agent’s scope while the rest of the fleet keeps running, and the trip itself becomes an incident record. Design the breaker per agent scope before you need it and test the trip in staging. Generated from the Body of Knowledge.
An agent receives a short-lived, scoped credential from its registry entry at deploy time, and every downstream call is verified and attributed in the audit log. No registry entry, no credential, no access. Generated from the Body of Knowledge.
For the actions the policy marks as needing a human, the agent pauses, a named reviewer decides within a time box, and the decision is logged as evidence. Define which actions need a human by policy, not by habit. Generated from the Body of Knowledge.
Discovery scans the places AI hides, matches each finding against the registry, and turns unknowns into registry entries or blocks, leaving a discovery report. Run it before you claim your inventory is complete. Generated from the Body of Knowledge.
A procured model or tool enters the inventory only after a gate that checks its documentation, evals and contract terms, and the outcome is recorded with any conditions. The gate is where deployer duties start, so keep its checklist versioned. Generated from the Body of Knowledge.
A use-case request becomes a structured record, passes the prohibited-use screen, gets a computed tier and switches on the gates that tier requires before the registry entry exists. Put every new AI request through this one path. Generated from the Body of Knowledge.
Threats are enumerated per data-flow element with STRIDE, extended with AI-specific attack classes, and each one is mitigated with a named test or accepted by a named owner. Fail the design review on any threat without a test. Generated from the Body of Knowledge.
Every training source enters a ledger row with its acquisition channel, licence or legal basis and reservation check, and lineage joins the rows to the models they trained. Query it when rights change, instead of retraining everything. Generated from the Body of Knowledge.
A dataset version is checked and signed into an admission record, and a read-time policy lets a job read it only for an admitted use, logging each read in lineage. Make every training job present its use case before it reads. Generated from the Body of Knowledge.
A fairness policy fixed before the run drives group and intersectional metrics, a proxy scan and a counterfactual flip test, gated on intervals and repeated on live decisions. Write the policy first, then let the suite fail the build. Generated from the Body of Knowledge.
Each consequential decision writes an explanation record with model, method and reason codes pinned, which is tested for fidelity and reused for notices, access requests and appeals. Answer every explanation duty from the same record. Generated from the Body of Knowledge.
The build signs a manifest of every model file and attaches provenance, the registry holds digest and signer, and serving verifies both before it loads a model. Refuse any load that does not match its registry entry. Generated from the Body of Knowledge.
Each public claim becomes a register row bound to the eval run behind it; rules check scope and interval before publication, and each release revalidates or pulls the claim. Register a claim before it reaches the copy. Generated from the Body of Knowledge.
The decision is recorded with its reason codes, the notice is generated from that record, and a contest reaches a reviewer who can change the outcome and logs it. Tie every notice to a decision record before you write the template. Generated from the Body of Knowledge.
One request is located through a data map, fanned out to a handler per store, answered in the weights by suppression now and retraining later, and closed with a fulfilment record. Build the data map before the first request arrives. Generated from the Body of Knowledge.
Staff reach approved AI tools through one gateway that applies the acceptable-use card by data class and emits a decision event per call, while discovery finds the tools used around it. Make the gateway the fastest route. Generated from the Body of Knowledge.
A change moves from a registered plan through shadow, pilot and canary to general availability, and a tripped criterion rolls it back to the pinned baseline. Register the criteria before the stage starts. Generated from the Body of Knowledge.
Live traffic, late labels and feedback by group are checked against thresholds; a breach pages an owner or opens an incident, and every check writes an evidence record. Give every signal an owner before you give it a chart. Generated from the Body of Knowledge.
Intended and prohibited uses as a Policy Card, a register that grants each consumer access for its approved use, and a monitor for use outside it. Register consumers before they get the outputs. Generated from the Body of Knowledge.
Registry facts and triggers feed a duty matrix, notices are drafted from versioned templates and approved by one owner, and every disclosure and notice sent is recorded. Generate the transparency page from the registry, not by hand. Generated from the Body of Knowledge.
A threshold or legal trigger goes to a named owner, evidence is frozen first, then a degraded mode, a regional switch or the full stop, and retirement ends in a recorded runbook. Drill the switches before a trigger fires. Generated from the Body of Knowledge.
The sequence an AI governance engineer runs across the workflows it owns, from intake and inventory through evals, gates and runtime to assurance evidence. Generated from the Body of Knowledge.
The five levels across the five stack layers, with an illustrative profile whose weakest layer sets the overall level. Assess each layer separately, then read the floor. Drawn from chapter 07.
The five maturity levels from Documented to Continuous, each proven by what the running systems show, with the trap that stalls the climb between them. Generated from the Body of Knowledge.
The serious-incident reporting windows for high-risk systems, by incident class, and the pipeline that meets them. Your pipeline must classify the incident before it can report. Drawn from chapter 08, as of 2026-09-24.
Who enforces each regime and how high its penalties can go: the AI Office for GPAI providers, national market-surveillance authorities for high-risk systems and the state Attorney General under California's SB 53, as of 2026-09-24. Record for each system which authority asks, so its evidence is ready for that authority. Drawn from chapter 08.
How a regulatory obligation flows to the engineering artefact that satisfies or supports it and the stack layer that artefact lives in: the reverse index made visible. Generated from the Body of Knowledge.
How the decisions that cannot be automated reach the committee and come back as data: triggers route a use case to the committee, its exceptions go into a register the policy gate reads, and the verdict shows the release passed under an exception until it expires. File exceptions as data with an expiry, not as minutes. Drawn from chapter 12.
The board sets appetite, the committee decides what the gates cannot, the second line builds the gates the first line runs through, and the third line re-performs gate decisions from the same evidence the board reads as KPIs and KRIs. Name who holds each role for your AI systems, and check that internal audit stays independent of what it assures. Drawn from chapter 12.
The four steps of the risk loop with the NIST AI RMF functions each covers and the stack layers that do its work, all under GOVERN and all writing to one risk register. Find the step your function skips, then build the layer that does its work. Drawn from chapter 13.
The illustrative five-by-five likelihood-by-severity matrix, with catastrophic severity on its own track as Critical at any likelihood, and the gate and acceptor each band triggers. Rate each risk on the defined scales, then wire each band to its gate. Drawn from chapter 13.
The five rungs of the mitigation hierarchy, worked top down, each with the stack control that implements it and the evidence it leaves; transfer sits beside the ladder, not on it. Start at the top and record why each higher rung was infeasible before you settle on a lower one. Drawn from chapter 13.
Where AI harm lands, from one person to the physical environment, with one example harm per level from the harms atlas and the control that catches it, coloured by its stack layer. Calibrate your severity scale against every level, not only against harm to one person. Drawn from chapter 13.
Provenance says where a dataset came from and on what terms; lineage traces how it moved through pipeline jobs and runs, backward from a model and forward from a dataset. Record lineage forward as well as back, because an erasure request or a licence withdrawal asks which models used the data. Drawn from chapter 14.
The build from intake to release as a chain of gates, each reading a structured record and each named with the pattern that implements it, ending in a technical file the pipeline compiles from those records. Find the first stage whose gate cannot refuse a missing record, and wire that one first. Generated from the Body of Knowledge.
The one control each combination of model type and deployment option adds on top of its row and its column, from classic predictive to open-weight models and from cloud hosting to an agentic wrapper. Find your combination and check that its control is in place before the go-live review. Drawn from chapter 15.
Explanation techniques on two axes, scope (global or local) and access (model-agnostic or model-specific), and the tests an explanation passes before its record is kept as evidence. Pick the quadrant your decision needs, then test the method for fidelity and stability before you rely on it. Drawn from chapter 16.
The first-report deadlines that one event can start at once under eight regimes, on one axis from awareness, as of 2026-09-24. Hold every clock in one incident record and alert on the nearest deadline. Drawn from chapter 17.
One tool call through the agent control plane: a registry entry gates the credential, the gateway and the guardrail check the call, a checkpoint fires where the stakes demand it and a per-agent breaker can stop it, while every step leaves telemetry kept as evidence. Check that each agent in your registry has all of them and that its stop has been drilled. Illustrative, not a claim of conformity. Drawn from chapter 23.
When each family of EU AI Act obligations applies, from the entry into force in 2024 to the legacy public-authority deadline in 2030, one lane per family, as of 2026-09-24. Find the lanes your systems sit in and plan the evidence for the next date on each. Drawn from chapters 18 and 08.
The questions that tell which EU AI Act operator roles an organisation holds for one system, each role with its core duties and the evidence it produces, and the Article 25 loop that turns a distributor, importer or deployer into the provider. Record the roles per system in the registry and ask the questions again whenever an Article 25 trigger fires. Drawn from chapter 18.
The four rungs the EU AI Act puts AI systems on by intended purpose, with the two high-risk routes and the Article 6(3) filter, the separate track for GPAI models, and side notes on what Korea, Texas, California and Colorado do instead, as of 2026-09-24. Place each system on every rung it meets, then read off its duties and dates. Drawn from chapters 18 and 21.
Twenty jurisdictions as equal tiles in rough geographic order, shaded by how binding their AI-specific regime is, with the US states and New York City in an inset, as of 2026-09-24. Find each jurisdiction you operate in, then read its section for the trigger, the clock and the enforcer. Drawn from chapter 21.
AI governance instruments in rising order of force, from principles and soft law through voluntary and harmonised standards to the treaty and the law, with the links the chapter traces between them, as of 2026-09-24. Build each control once and tag it with every instrument it serves, because standards support and do not confer. Drawn from chapter 22.