The one-document lab: one AI system, three regimes, zero duplicate work
The short answer
The crosswalk is the second payoff of Module 6
Where Topic 6.6 produced a decision about where a feature may ship, this topic produces the artifact that keeps your evidence for that feature, and every other feature, consistent across every regime that asks about it.
What you will be able to do
- Create a single evidence crosswalk document that maps one AI system's existing artifacts (the conformity file, the AI systems inventory, the data provenance file, the evaluation report) to the requirements of three distinct regimes at once: the EU AI Act's Annex IV technical documentation, ISO/IEC 42001's Annex A controls and clauses, and a Singapore MGF and MAS materiality file.
- Distinguish evidence reuse (gathering a fact once and citing it in three places) from regulatory equivalence (one regime's clearance satisfying another regime's requirement), and explain why the second is a defect the first is designed to avoid.
- Trace each of the EU AI Act's nine Annex IV technical documentation slots to the ISO/IEC 42001 Annex A control objective and the MAS AI Risk Management Guidelines (AIRG) section that ask a related question of the same underlying evidence.
- Assemble a materiality assessment and AI inventory entry consistent with the MAS AIRG's proposed four sections (oversight, risk management systems, life cycle controls, capabilities and capacity) for a system you already documented for the EU AI Act.
- Label every Singapore instrument correctly by its legal status (the PDPA is statute; the PDPC's generative AI guidelines and the MGF family are advisory or voluntary; the MAS AIRG is proposed and finalization-pending as of 15 August 2026) inside a single document that also cites binding EU law, without letting the strongest regime's tone bleed into the weakest one's description.
- Identify the genuine gaps: the small number of questions each regime asks that no other regime asks at all, so the crosswalk shows what is truly shared and what is truly regime-specific rather than pretending everything reduces to one universal form.
- Produce three thin, regime-specific cover packages, each pointing back into the single evidence crosswalk rather than restating its contents, so a reader in Brussels, an auditor from an accredited certification body, and a Singapore compliance officer can each open a document written in their own regime's language.
- Defend the crosswalk against the two hardest challenges a skeptical reviewer will raise: that reusing evidence across regimes hides shortcuts, and that a document built for three audiences at once satisfies none of them well.
The ideas, one by one
Facts are shared; questions and conclusions are not
Gather each underlying fact once, dated and versioned, and cite it from three regime-specific cells. Never let one regime's conclusion (a MAS materiality score) silently answer another regime's different question (the EU AI Act's Article 9 necessity case).
The nine-row crosswalk aligns EU AI Act Annex IV, ISO/IEC 42001 Annex A, and the MAS AIRG's four sections
by the underlying question each is really asking, not by matching document titles. Building it, row by row, for a real system is the Create-level skill this topic teaches.
Name the residue honestly
The EU declaration of conformity and CE marking, ISO's accredited audit cycle, and MAS's non-delegation doctrine for third-party AI are each genuinely regime-specific. A crosswalk that forces every row into a false three-way equivalence is less trustworthy than one that marks its own gaps.
Singapore's legal weight must stay exact inside a mixed document
The PDPA is statute; the PDPC's guidelines are advisory; the MGF family is voluntary; the MAS AIRG is proposed and finalization-pending as of 15 August 2026. Letting the EU AI Act's binding tone bleed into a Singapore cell misstates the law.
The deployment-versus-governance gap is real and measured
Deloitte found 72 percent of Singapore businesses plan agentic AI deployment within two years against only 14 percent reporting mature agentic governance; AWS-commissioned Strand Partners research found under 30 percent of AI-adopting Singapore SMEs have a clearly accountable owner. This is the market condition the crosswalk exists to answer.
Cite, never restate
Every factual sentence in each of the three cover packages must trace to a specific, dated crosswalk row. A restated fact is a fact that can silently drift from the source it was supposed to match.
The crosswalk is alive, not filed once
Every row needs a review trigger, and Singapore's instruments move fastest in this window (the MAS AIRG finalizing, the Agentic MGF's version updates), so the crosswalk's Singapore cells need the tightest watch.
Evidence reuse is not regulatory equivalence
No certificate, declaration, or materiality file from one regime substitutes for another's. The crosswalk eliminates duplicate labor in gathering facts; it never eliminates the separate judgment each regime demands.
Build every row to survive the toughest reviewer it might face
An ISO auditor's live document request, an EU authority's post-incident foresight question, and a Singapore examiner's proportionality conversation each test a crosswalk row differently; a row built for the toughest of the three passes the others without extra work.
Listen to it
The same lesson, talked through by two hosts as episode 50 of the podcast.
Real cases
These examples show the underlying instruments and the market conditions the crosswalk skill responds to. None of them is fictional; each is cited to its primary or reputable source.
Example 1: The governance-maturity gap the crosswalk exists to close (Deloitte, 2026). Deloitte AI Institute's "The State of AI in the Enterprise: The Untapped Edge," published 3 February 2026 from an August to September 2025 survey of 3,235 leaders in 24 countries including 75 in Singapore, found 72 percent of Singapore businesses plan to deploy agentic AI within two years against only 15 percent doing so today, while just 14 percent report a mature model for agentic AI governance, below the 21 percent global average. The number is the reason this lab is worth building: deployment is outrunning documentation across the whole market, and an organization that can hand a regulator or a buyer a single, coherent evidence base is operating well ahead of its peers rather than merely keeping pace with them.
Example 2: The oversight gap underneath the deployment (AWS-commissioned Strand Partners, 2026). The AWS-commissioned Strand Partners research, published 6 May 2026 from 1,500 Singapore businesses across financial services, healthcare, and manufacturing, found just under 30 percent of AI-adopting SMEs have a clearly defined person responsible for overseeing AI accuracy, and just under 40 percent have no formal escalation process when an employee is unsure about an AI output; six in ten would face significant or moderate disruption if the person primarily responsible for AI left, with about one in ten expecting AI initiatives to stop altogether. Row 8 of the crosswalk, accountability, exists precisely to close this gap: naming the accountable owner once, consistently, across every regime's document, is a direct answer to a documented market weakness, not a theoretical exercise.
Example 3: The instrument that shows a regulator itself building crosswalk thinking (IMDA's Agentic MGF, 2026). The Model AI Governance Framework for Agentic AI reached version 1.0 on 22 January 2026 at the World Economic Forum and version 1.5 on 20 May 2026 (updated 5 June 2026), with the update adding case studies and best practices, IMDA's own move toward showing organizations how to apply the four dimensions concretely rather than leaving the mapping work entirely to industry. (see Topic 6.7) This is worth noticing because it means the regulator expects the kind of concrete, evidenced mapping this lab teaches you to build, not a restatement of the framework's four dimensions in your own words.
Example 4: The certifiable instrument's real-world proof point (Anthropic, ISO/IEC 42001, 2025). Anthropic's accredited ISO/IEC 42001:2023 certification, issued by Schellman and announced 13 January 2025, is the concrete demonstration that the ISO column of the crosswalk is not a hypothetical exercise; an accredited certification body genuinely audits the evidence a Statement of Applicability points to. (see Topic 6.3) The lesson for the crosswalk: an ISO auditor reads exactly the kind of pointer-to-dated-evidence structure this lab teaches, because that structure is what an audit is designed to verify quickly.
Example 5: The instrument that proves "voluntary" still carries real supervisory weight (MAS AIRG doctrine). The MAS AIRG Consultation Paper (P017-2025, 13 November 2025, consultation closed 31 January 2026, final not issued as of 15 August 2026) states as doctrine that proportionality governs the degree of implementation, not whether the guidelines apply at all, and that a financial institution may not delegate governance responsibility for third-party AI to the vendor supplying it. (see Topic 6.8) Neither statement is a formality; both are precisely the kind of expectation a materiality file must answer directly, which is why Part 2 and Part 4 of the Singapore column in 3F exist as named parts rather than an afterthought.
Example 6: A published standard's own instruction to reuse (ISO/IEC 42005:2025). ISO published a dedicated impact-assessment guidance standard, ISO/IEC 42005:2025, whose Annex A maps back to ISO/IEC 42001, so that an organization that has already done impact-assessment work is formatting existing work into the shape an ISO auditor expects rather than starting from zero. (see Topic 6.3) This is the exact reuse principle the crosswalk generalizes across regimes; ISO built it into its own standards family for a reason, and the reason is the same one driving this whole lab: duplicate elicitation of the same fact is waste, and every serious governance instrument eventually says so in its own way.
Example 7: The Vietnam contrast that shows legal-weight labeling has real teeth (Law No. 134/2025/QH15). Vietnam's Law on Artificial Intelligence, passed by the National Assembly 10 December 2025 and in effect since 1 March 2026, is a binding statute, in sharp contrast to every Singapore instrument in this lab. (see Topic 6.9) A crosswalk extended to cover a Vietnamese deployment must label that column's obligations "requires," matching the EU AI Act's tone, directly beside a Singapore column correctly labeled "voluntary" or "proposed." The two columns sitting side by side, correctly labeled, is the clearest proof that legal-weight labeling is not a stylistic nicety; get it backward and a Singapore cell reads as binding law that does not exist, or a Vietnam cell reads as optional guidance that a court can actually enforce.
Example 8: The transition period as a built-in review trigger (MAS AIRG's proposed twelve-month window). The MAS AIRG's consultation paper proposes a twelve-month transition period after the final guidelines are issued, before full compliance is expected. (see Topic 6.8) A crosswalk built now, with the Singapore cells correctly marked "proposed," has a natural review-trigger structure already built into the instrument itself: the trigger is not a vague "check back sometime," but two concrete, named dates, the day the guidelines are issued and the day the transition period ends, both worth entering directly into the crosswalk's row-level triggers the moment they are known.
Where people go wrong
- "One certificate satisfies every regime." ISO/IEC 42001 certification does not make a system compliant with the EU AI Act, and neither ISO certification nor EU AI Act conformity satisfies Singapore's MAS or MGF expectations. Each regime asks its own question of the shared evidence and reaches its own conclusion; the crosswalk saves labor on gathering facts, never on the judgment each regime demands.
- "If we already scored the materiality assessment, the EU necessity case is done." A moderate MAS materiality score answers a supervisory-soundness question, not the EU AI Act's fundamental-rights necessity question under Article 9. The two conclusions can differ even when they draw on the same facts, as Vernon's near-miss on row 5 shows. Write both, and never let one regime's conclusion silently stand in for another's.
- "The crosswalk is a fourth document to maintain, so it adds work." Reversed. The crosswalk replaces three separately drifting documents with one dated source of truth that each cover package cites; maintaining one crosswalk with named review triggers is less labor than maintaining three documents nobody has told to talk to each other, and it is the only way to guarantee the three cover packages stay consistent with each other over time.
- "Every row must map cleanly across all three regimes, or the crosswalk has failed." Some rows genuinely do not line up (the necessity case, the declaration of conformity), and naming that misalignment honestly is more useful than forcing a false equivalence. The residue each regime carries alone (Section 3C) is not a gap in the crosswalk; it is accurate information the crosswalk correctly surfaces.
- "Voluntary Singapore guidance can be summarized in the same confident tone as binding EU law." The MAS AIRG is proposed and finalization-pending as of 15 August 2026; the MGF family and the PDPC's generative AI guidelines are voluntary and advisory respectively. A crosswalk that lets the EU AI Act's binding tone bleed into the Singapore cells misstates the law and fails this program's honesty rules the moment a reader checks the citation.
- "Building the crosswalk once means it never needs updating." The Singapore column moves fastest in this lab's window (the AIRG finalizing, the Agentic MGF's version 1.5 update), so the crosswalk needs named review triggers, not a one-time build. A stale crosswalk that still looks authoritative is worse than an acknowledged gap, because a reader trusts it exactly when it can no longer be trusted.
- "A cover package should restate the evidence in full so it stands alone." A cover package that restates rather than cites is exactly how three documents drift apart: the restated version and the crosswalk's version inevitably diverge over time, and nobody notices until a reviewer compares them. Every factual sentence in a cover package should point at a crosswalk row, not repeat it from memory.
- "If procurement asks whether we are ISO certified, the honest but incomplete answer is worse than a vague yes." The honest answer, naming exactly what exists (a Statement of Applicability in progress, backed by a real evidence crosswalk) and what does not (a certificate not yet issued), is stronger than a vague or inflated yes, because it is verifiable and it demonstrates the underlying discipline a sophisticated buyer is actually screening for.
- "The crosswalk only matters for systems that are already in scope of all three regimes." Building the crosswalk habit on any system that touches two or more regimes, even loosely, is what prevents duplicate elicitation from becoming the default the next time a third regime shows up. A system in scope of only Singapore's voluntary guidance today may gain EU reach the moment its output is used there. (see Topic 6.6) The habit, not the trigger, is what should generalize.
- "A closed consultation is close enough to a final rule to cite with confidence." The MAS AIRG's consultation closed 31 January 2026, and the final guidelines were still not issued as of 15 August 2026. A closed consultation means MAS is reviewing feedback, not that the text is settled; the substance of a final rule can differ from the consultation draft. Cite the consultation paper as what it is, a proposed instrument under review, and update the crosswalk the day final guidelines are actually issued, not the day the consultation window closed.
- "Since the crosswalk is an internal document, its legal-weight labeling matters less than the external cover packages." The internal crosswalk is what every cover package is generated from; a mislabeled row in the internal document propagates its error into every package built from it. Getting the label right once, at the source, is what makes every downstream document correct automatically; getting it wrong once, at the source, guarantees every downstream document repeats the same mistake.
Questions people ask
- What is evidence crosswalk?
- A single, dated document that maps one AI system's underlying facts to the requirements of multiple governance regimes at once, one row per underlying question, with regime-specific cells that cite the shared fact rather than restate it. It is the artifact this topic teaches and the second payoff artifact of Module 6.
- What is evidence reuse versus regulatory equivalence?
- The core distinction the topic teaches. Evidence reuse means gathering a fact once and citing it in multiple regime-specific documents. Regulatory equivalence would mean one regime's clearance (a certificate, a declaration) satisfying another regime's separate requirement, which is not true of any pairing in this lab; each regime's judgment must still be reached on its own terms.
- What is regime-specific residue?
- A requirement genuinely unique to one regime with no counterpart cell in the crosswalk's other columns, such as the EU AI Act's Article 47 declaration of conformity, ISO/IEC 42001's accredited certification audit cycle, or the MAS AIRG's proposed non-delegation doctrine for third-party AI. Marking residue honestly, rather than forcing a false equivalence, is part of what makes a crosswalk defensible.
- What is cover package?
- A short, regime-specific document (the EU AI Act technical documentation, the ISO/IEC 42001 Statement of Applicability extract, the Singapore MGF and MAS materiality file) that draws entirely on the evidence crosswalk, citing its rows rather than restating their contents.
- What is EU AI Act Annex IV technical documentation?
- The nine-slot documentation structure a provider of a high-risk AI system must draw up under Article 11, covering the system's general description, development process, capabilities and limitations, performance metric justification, risk management under Article 9, lifecycle changes, standards applied, declaration of conformity, and post-market monitoring plan. (see Topic 5.6)
Keep going
This lesson builds Framework crosswalking without false equivalence, and that page shows the roles that hire for it. Every Certified AI Governance Professional (CAIGP) lesson.
Watch the lesson and prove it
Proving it happens inside the program: watch, explain it in your own words, and it goes on a record employers can check.
Watch the lesson and prove it →