The reorganization memo that finally worked did not start in the localization department. It started in a legal deposition. A global medical-device company had shipped a translated instructions-for-use document in which a fluent, grammatical machine rendering had inverted a contraindication, and the question the general counsel asked in the post-incident review was not "which vendor made the mistake." It was "who in this company owned the decision that this content was safe to machine-translate, and who owned the sign-off that it passed." The honest answer was nobody. Localization sat three levels down under marketing operations, staffed by a manager whose title and compensation were pegged to words shipped per month and cost per word driven down quarter over quarter. The org chart had been optimized for a world where humans produced the words and volume was the constraint. In that world the chart was rational. In the world that had actually arrived, where an engine produced every first draft in seconds and the scarce, expensive, unautomatable work was owning whether the output was correct, terminology-faithful, and safe, the chart was pointed at exactly the wrong things. This lesson is about redrawing it. Not with new boxes for their own sake, but by moving authority to where the risk and the value now live: onto a small number of quality-ownership roles that the machine's arrival created and that most enterprise localization functions have not yet built. We will define each new role, its mandate, what it owns, where it sits, and the reporting lines and authority that make it real, and we will end on a worked, concrete AI-enabled org chart you could adapt on Monday.

Why the Org Chart Has to Change

An org chart is a bet about where the hard, valuable, failure-prone work lives. You put the senior people, the authority, and the money where the bet says the value and the risk concentrate. For most of the localization industry's history that bet was simple and correct: the value lived in producing words at quality and volume, so the chart was a production pyramid. Translators produced, revisers checked, project managers moved files and schedules, and a localization manager owned throughput and cost. The unit was the word, the constraint was human production capacity, and every box on the chart was ultimately a lever on producing more words at an acceptable standard for less money.

Then the constraint moved. By 2024, machine-translation post-editing (MTPE), the workflow in which a human edits engine output rather than translating from a blank page, had reached roughly 46% adoption, up from about 26% in 2022, and 81% of language-service providers (LSPs), the agencies that sell translation as a service, offered it. A hybrid workflow lifted a linguist from around 2,000 words a day to 5,000 or more, and MTPE priced at roughly 50 to 75% of full human translation. When the engine took over first-draft production, human production capacity stopped being the binding constraint on the operation. Volume became cheap. And the moment volume becomes cheap, an org chart optimized to squeeze more volume out of humans is optimizing the wrong variable. It is spending its senior authority and its compensation bands on a problem the machine already solved, while the problem the machine created goes unowned.

The problem the machine created is specific. Engine output is fluent first and accurate second: grammatical, confident, natural-sounding prose that can mean the opposite of the source and that no automatic score reliably flags. Studies of large-language-model output on medical content found error rates around 59% on drug names, 60% on dates and times, and 66% on adverse events, every one delivered in perfect prose. The scarce, expensive, unautomatable work is now owning the gap between fluent and correct: deciding which content the engine may touch and which it must never touch, holding the approved terminology against an engine that drifts off it, scoring output against a severity typology so a single dangerous error blocks delivery, and standing behind all of it to a client's auditor or a regulator. That work does not scale down as volume scales up. It scales up, because every additional machine-drafted segment is another place a silent critical error can hide. The org chart has to move its authority and its money onto that work, which means it has to name and empower the roles that do it.

An org chart is a bet about where the hard work lives. When the engine took over production, the hard work moved from producing words to owning whether the words are correct, terminology-faithful, and safe. A chart still betting on volume is spending its authority on the one thing the machine already does.

The Forcing Function Is Not Optional

Two standards make this a governance requirement, not a philosophy. The revised ISO 18587, the international standard for post-editing machine translation output (in DIS ballot, publication targeted for late 2025 into 2026), expands its scope from machine translation to "non-human translation output," explicitly covering AI and large language models, and insists the person who signs off on post-edited content hold the same full professional-translator competence as a human translator. It retires the rigid light-versus-full split for an effort spectrum and puts the accountability squarely on a qualified human. ISO 5060:2024 formalizes the MQM-aligned Critical / Major / Minor error scoring that decides whether output ships. Together they mean that "the engine wrote it" is never a defensible answer when a dangerous error ships, and that a qualified human must own the quality decision on the record. An org chart with no role that unambiguously owns that decision is an org chart that cannot be conformant. The reorganization is not an HR preference. It is what the standards, the liability, and the economics jointly require.

The Head of Localization AI

The first new role is the one the deposition was missing. The Head of Localization AI is the single accountable executive owner of the machine-enabled localization operating model: the person who owns the engine and vendor strategy, the quality-governance program, the risk-tiering policy that decides what the engine may and may not touch, and the profit-and-loss consequences of getting any of it wrong. This is not a rebranded localization manager who added "AI" to a business card. It is a role defined by a mandate the old chart never assigned to anyone: own the operation as an AI-native, human-owned-quality system, and be the one name in the room when leadership, legal, or a regulator asks who is accountable for it.

The Mandate

The mandate has a shape worth stating precisely, because a vague mandate is why the old localization manager could not have absorbed this. The Head of Localization AI owns three things that must sit together in one accountable person and that the production-era chart scattered or left unowned. First, the operating model: the decision that the operation runs MT-first with humans on quality and judgment, and the design of the pipeline that makes that safe rather than reckless. Second, the risk posture: the enterprise policy for which content is MT-eligible under which post-editing effort, and which content is MT-forbidden, held consistently across every language, content type, and engine. Third, the accountability: being the executive who can tell the board "here is our throughput, here is the risk tier each content type received, here is our error rate with zero shipped Criticals in regulated content, and here is why it is defensible under ISO 18587 and 5060." The reason these three cannot be split is that each one is a lever on the others. You cannot own the throughput without owning the risk it creates, and you cannot own the risk without owning the model that produces it.

What the Role Owns

• Engine and vendor strategy: which MT and LLM engines the operation uses, on what data terms, evaluated on domain quality, terminology adherence, and critical-error rate rather than a headline benchmark, and the concentration-risk posture that prevents the operation from becoming hostage to a single engine.
• Quality-governance charter: the standing authority of the quality function, including its power to fail a delivery, and the operationalization of ISO 18587 and ISO 5060 into an org-wide program rather than a slogan.
• Risk-tiering policy: the enterprise rule set that classifies content by consequence and routes it, so a drug label never lands on the cheap workflow by accident.
• Linguistic-asset strategy: the translation memories and termbases as governed enterprise assets, because they are the operation's durable competitive advantage and the thing that grounds the engine on approved language.
• The commercial story: pricing defensible quality tiers to clients or justifying the operation's cost and risk posture to internal leadership, selling provable quality instead of racing competitors to the bottom on raw machine output.

Where It Sits and Reports

Placement is the whole point of this role, and getting it wrong wastes the role. The reason the deposition happened is that localization sat buried under marketing operations, where its charter was cost and volume and it had no standing to say no to a risky machine-translation request. The Head of Localization AI has to sit high enough that its risk authority is real. In practice that means reporting to a function that owns enterprise risk or product quality rather than to a function that owns marketing throughput: commonly a VP of Product, a Chief Content or Chief Digital Officer, a Head of Global Operations, or in regulated industries a peer relationship with legal, quality, and regulatory affairs. The test is not the title's altitude for its own sake. The test is whether this person can, without escalating past their own authority, refuse to ship content that is not safe. If the answer is no, the role is decorative and the deposition will happen again.

The Head of Localization AI is defined by one power the old localization manager never had: the standing authority to refuse to ship content that is not safe, without escalating past their own seat. Place the role wherever that authority is real, and nowhere lower.

The Quality Lead

If the Head of Localization AI owns the model, the Quality Lead owns the gate. The Quality Lead is the senior individual accountable for the operation's quality-evaluation program: the person who owns the error typology, the severity-scored quality gate that decides whether a file ships, the qualification of the people who evaluate, and the defensible quality record that proves conformance. This is the role that turns the program's central discipline, the severity-scored gate that catches the silent critical error, from a workflow step into an owned function with a name and authority behind it.

The Mandate

The Quality Lead's mandate is to make quality a measured, defensible property of every shipped file rather than a vibe. That means owning the analytic evaluation framework the operation scores against, MQM, Multidimensional Quality Metrics, which classifies every error by dimension (accuracy, terminology, locale conventions, fluency) and by severity, as formalized for translation output in ISO 5060:2024. It means owning the rule that decides the go/no-go: a single Critical error, one that renders content dangerous, legally exposed, or actively misleading, blocks delivery no matter how clean the rest of the file reads, while Major errors (meaningful degradation) and Minor errors (cosmetic flaws) are counted and weighted. And it means owning the artifact the gate produces: a structured error record, counted by dimension and severity, that a client's auditor or a regulator can read and trust. The Quality Lead is the person who can stand behind the sentence "this file passed" and defend exactly what that means.

What the Role Owns

• The evaluation program: the MQM/ISO 5060 typology as the operation applies it, the severity definitions calibrated to the operation's content, and the sampling and scoring methodology that make evaluations repeatable across evaluators.
• The delivery gate: the one-Critical-fails rule and the authority to enforce it, including the authority to block a delivery that a project manager or a client wants shipped on schedule.
• Evaluator qualification: who is allowed to score, how they are calibrated so two evaluators reach the same severity on the same error, and the ongoing checks that keep them aligned.
• The quality record: the audit-grade, segment-level documentation of source, edit, term decision, and error score that proves conformance and preserves the human accountability the revised ISO 18587 requires.
• The distinction between the score and the verdict: owning the operation's discipline that an automatic quality-estimation (QE) confidence number, a model's reference-free guess at whether a segment is fine, routes human effort but never ships anything. The verdict is always a human evaluation, because a confidence score can rate the silent critical error as safe.

Where It Sits and the Independence Problem

The Quality Lead reports to the Head of Localization AI, but the reporting line carries a structural tension that has to be designed around deliberately. Quality's job is to fail files. Production's job is to ship them. If the Quality Lead reports through the same manager whose bonus depends on on-time delivery, the gate will quietly bend under schedule pressure until it is theater. The design that works separates the two lines: the Quality Lead's authority to fail a delivery is independent of the production manager's authority to schedule it, and both roll up to the Head of Localization AI, who is accountable for the trade-off. In a mature operation the Quality Lead has a dotted line to enterprise quality or regulatory affairs precisely to reinforce that independence, so that the gate cannot be overruled by a delivery deadline alone. The analogy that lands with executives is financial audit: you do not let the people who book the revenue also sign off that the books are clean, and for the same reason you do not let production own the quality gate.

Quality's job is to fail files and production's job is to ship them. If the Quality Lead reports through the person whose bonus depends on shipping, the gate becomes theater. Independence is not a nicety here; it is the whole function.

The Terminology Lead

The third new role owns a failure mode that is invisible on any throughput dashboard and lethal on a regulatory one. The Terminology Lead is the senior owner of the operation's terminology as a governed enterprise asset: the person accountable for the termbases, the controlled glossaries of a client's or the enterprise's approved terms with their correct equivalents in each language, and for the policy that makes those approved terms hold through machine translation and post-editing across every project, from intake to delivery.

Why Terminology Needs Its Own Owner

It is tempting to fold terminology into quality, and in a small operation the same person may wear both hats. But at enterprise scale terminology is a distinct discipline with a distinct failure mode, and the machine sharpened it. An engine optimizes for the most probable fluent target, and it neither knows nor cares that a device maker calls a specific component "the cannula" and has spent a regulatory submission ensuring every document in every language uses that exact word. The engine reaches for whatever near-synonym it saw most often in training, "the needle," "the tube," renders a perfectly clean sentence with the wrong approved term, and the drift is invisible because nothing about the sentence looks broken. Worse, an unverified machine output using the wrong term gets written into the translation memory (TM), the database of approved source-target pairs the tools reuse, and now the error propagates: every future project that leverages that memory inherits it, multiplied across languages and releases. Terminology fidelity is a control the engine cannot supply for itself, it decays silently, and it compounds. That is exactly the profile of a risk that needs a named owner rather than a shared afterthought.

What the Role Owns

• The termbases as governed assets: their structure, their approval workflow, their integration into the CAT tool (the computer-assisted translation environment linguists work in) and the TMS (translation-management system) so the engine and the post-editor are flagged the moment they depart from an approved term.
• Terminology policy: how approved terms are extracted (increasingly with AI assistance, then verified against the source domain before becoming a rule), defined with subject-matter experts, and enforced consistently across an account or the whole enterprise.
• Grounding the engine: ensuring the operation's engines and LLMs are grounded on the approved termbase and TM rather than the open web, so the machine answers from the enterprise's approved language.
• TM hygiene: the governance that keeps the translation memory a compounding asset rather than an infection vector, so leverage multiplies quality instead of multiplying a term error.
• Vertical depth: the domain expertise in the operation's high-value verticals (medical, legal, financial, technical) that makes a term choice defensible to a client's subject-matter experts and regulators.

Where It Sits and What It Ranks

The Terminology Lead reports to the Head of Localization AI, peer to the Quality Lead, and the two coordinate constantly because a terminology error is often scored as an accuracy or terminology error at the quality gate. The placement decision that matters is seniority: in a production-era chart terminology was frequently a part-time task handed to whichever linguist cared most, with no authority to enforce anything. In the AI-enabled chart it is a lead role with the authority to set enterprise terminology policy and to require that engines be grounded on the approved termbase, because in regulated content a term error is not cosmetic. It can be the Critical that fails the file and the drift that a regulator cites. Under-ranking this role is one of the most common and most expensive mistakes an operation makes when it redraws the chart, because the cost of getting terminology wrong is invisible right up until it is enormous.

How the Chart Shifts from Volume to Quality Ownership

Naming three roles is not a reorganization. The reorganization is the shift in where authority and money sit, and it is best understood as a set of migrations from old boxes to new ones. It helps to see them side by side, because leadership will ask "are we just adding headcount," and the honest answer is mostly no: this is a re-pointing of existing authority, not a pile of net-new hires.

The Migrations

• From "words per month" to "quality risk owned." The localization manager whose scorecard was throughput and cost-per-word becomes, or reports to, a Head of Localization AI whose scorecard is throughput and a defensible quality-risk posture. The metric moves from a single volume axis to a dual axis, and the senior authority follows the second axis, because that is the one the engine did not automate.
• From "reviser as a production step" to "Quality Lead as an independent gate." Revision used to be a task inside production, subordinate to the schedule. It becomes an owned function with the authority to fail a delivery, structurally separated from production so the gate cannot be bent by a deadline.
• From "terminology as a part-time favor" to "Terminology Lead as a governed asset owner." The glossary that used to live in a spreadsheet maintained by whoever had time becomes a governed enterprise asset with a named owner and enforcement authority.
• From "translator seats" to "post-editing and judgment seats." The largest population, the linguists, does not disappear, but its center of gravity moves from producing words to post-editing engine output, evaluating it, and owning the judgment calls. The headcount may even hold roughly flat while its composition and its compensation logic shift from volume to quality ownership.
• From "PM owns delivery" to "PM owns delivery within a risk-tiered, gated model." The project manager still owns the schedule and the client relationship, but now quotes a defensible quality tier instead of racing to the bottom on price, and delivers within a gate they do not control. The PM's job gets harder and more valuable, not obsolete.

The through-line is that the pyramid inverts its emphasis. The production era put its scarce senior authority at the top of a volume pyramid and its money in the width of the base. The AI-enabled era puts its scarce senior authority on the quality-ownership roles, the Head of Localization AI, the Quality Lead, the Terminology Lead, and keeps a strong base of post-editors and engineers whose value is now their judgment rather than their typing speed. Nobody's job vanished in a puff of automation. The authority moved onto the work the machine made scarce, and the compensation logic moved with it.

This is a re-pointing of authority, not a hiring spree. The same three questions the deposition exposed, who decided this was safe, who owns the gate, who owns the terminology, become three named roles with real power, and the money follows them.

What Does Not Change, and Why That Matters

It is worth being precise about what the reorganization does not touch, because over-rotating is its own failure. The localization engineer, the role that owns the structural failure modes, broken code placeholders, length overruns past a UI's pixel budget, encoding corruption, subtitle timing, does not become less important; if anything the engine's tendency to break structure raised the demand for it, and it keeps its place on the engineering side of the chart. The linguists remain the largest population and the source of the domain judgment everything else depends on. The client-facing PM function remains essential. The reorganization is not a purge of the production roles. It is the addition of a quality-ownership layer with real authority above and beside them, and the re-pegging of seniority and compensation to reward the judgment the machine cannot supply. An operation that fires its linguists and installs three leads with nobody to lead has not transformed; it has decapitated itself.

Hiring and Growing Into These Roles

The scarce resource in this whole reorganization is not the org chart. It is the people who can fill the three new roles, and in 2026 the supply has not caught up to the demand the engine created. The Dublin postings for these titles sit unfilled for months. So the operating question for a leader redrawing the chart is not only "what are the boxes" but "where do the people come from," and the answer is almost always grow-first, hire-to-fill-gaps.

Grow From Within First

These roles are layerings on top of deep linguistic competence, not lateral hires from outside the discipline, and that is a feature. The Quality Lead is a strong evaluator who has internalized the MQM dimensions and the severity calibration and has grown the judgment to run a program and qualify others. The Terminology Lead is a terminologist with vertical domain depth who has grown into policy ownership. The Head of Localization AI is most defensibly grown from a senior localization leader, a PM or quality manager, who has added the engine, governance, and risk fluency this curriculum teaches at its strategic tiers. Growing them from within has three advantages that hiring rarely matches: the domain and terminology knowledge is already there and is the hardest part to acquire, the institutional knowledge of the operation's content and clients is intact, and it solves the retention problem in the same move. The vendor pool and staff linguists quietly leaving the field because the work feels like cleaning up after a machine for less money are exactly the people these roles retain, by giving them a destination that moves them up rather than out.

What to Look For, and What to Avoid

The signal to hire or promote on is not words-per-hour and not tool certifications. It is demonstrated ownership of the gap between fluent and correct: a track record of catching the silent critical error, of defending a term choice to a subject-matter expert, of holding a quality standard against schedule pressure, of reading an automatic QE score skeptically rather than trusting it. The anti-pattern to avoid is hiring a generalist "AI leader" with no localization depth to run the Head of Localization AI role, on the theory that the AI is the hard part. It is not. The engine is a commodity; the judgment about what it may touch, whether its output is correct, and which terminology it must honor is the scarce part, and it is grounded in linguistic and domain competence a generalist does not have. The revised ISO 18587's insistence that the accountable human hold full professional-translator competence is not bureaucratic box-ticking. It is the standard codifying exactly this: you cannot own quality on machine output without the competence to judge it.

The Development Path, Made Explicit

• Foundation: deep linguistic and at least one vertical's domain competence, plus the AI-aware skepticism that spots the fluent-but-wrong output. Everything else layers on this.
• To Terminology Lead: add termbase governance, term-extraction tooling, TM hygiene, and the authority to set and enforce policy. Grown from a strong terminologist.
• To Quality Lead: add MQM/ISO 5060 fluency, severity calibration, evaluator qualification, and the quality-record discipline. Grown from a strong evaluator or reviser.
• To Head of Localization AI: add engine and vendor strategy, governance operationalization, risk-tiering policy, concentration-risk management, and the commercial and leadership fluency to own the P&L and defend the model to the board. Grown from a senior PM, quality manager, or localization leader.

A Worked AI-Enabled Org Chart

Abstract principles are easy to nod at and hard to act on, so here is a concrete chart for a mid-to-large enterprise localization operation running MT-first across a dozen or more languages and a mix of low-risk and regulated content. Adapt the headcount to your scale; the structure and the authority lines are the load-bearing part.

The Top of the Chart

At the top sits the Head of Localization AI, reporting to an enterprise function that owns risk or product quality, not marketing throughput, and holding standing authority to refuse unsafe content. This person owns the operating model, the engine and vendor strategy, the risk-tiering policy, the linguistic-asset strategy, and the P&L. Three functions report into this seat, and the design intent is that quality and terminology are peers to production rather than subordinate to it.

The Three Reporting Functions

• The Quality function, led by the Quality Lead. This function owns the evaluation program, the severity-scored delivery gate, evaluator qualification, and the quality record. Under the Quality Lead sit the QE analysts and evaluators who score output against MQM/ISO 5060. Critically, this line has the authority to fail a delivery and a dotted line to enterprise quality or regulatory affairs to keep that authority independent of the production schedule.
• The Terminology function, led by the Terminology Lead. This function owns the termbases as governed assets, terminology policy, engine grounding on approved language, and TM hygiene. Under the Terminology Lead sit terminologists with vertical domain depth. This function is peer to quality, not folded into it, because the failure mode it owns is distinct and compounding.
• The Production function, led by a production or delivery manager. This is where the volume gets processed and where most of the headcount lives: the MTPE leads who own post-editing workflows, the post-editors who clear the machine-drafted files, the localization PMs who own delivery and quote defensible tiers, and the localization engineers who own the structural plumbing (placeholders, length budgets, encoding, and the build pipeline). Production owns throughput and delivery, but it ships through the quality gate it does not control.

The Authority Lines That Make It Work

The boxes are the easy part; the lines are where operations succeed or fail. Three authority relationships make this chart function rather than merely exist. First, the gate line: the Quality Lead's authority to fail a delivery is independent of the production manager's authority to schedule it, and the two are reconciled only at the Head of Localization AI, who owns the trade-off between speed and safety. This is the single most important line on the chart, because it is the one the deposition exposed as missing. Second, the grounding line: the Terminology Lead's approved termbase is a required input to production's engines, not a suggestion, so the machine is grounded on approved language before it drafts. Third, the escalation line: when content is flagged as potentially MT-forbidden at intake, the routing decision escalates to a policy the Head of Localization AI owns, so no PM under schedule pressure quietly sends a drug label to the cheap workflow to hit a date. These three lines encode the whole thesis: humans own quality and judgment, the engine owns first drafts, and the authority to say no lives where it can actually be exercised.

Reading the Chart Against the Deposition

Return to the general counsel's two questions, because a good org chart is one that can answer them out loud. "Who owned the decision that this content was safe to machine-translate?" On this chart, the risk-tiering policy is owned by the Head of Localization AI and the intake escalation line ensures a human applied it; the decision has a name. "Who owned the sign-off that it passed?" The Quality Lead owns the gate, a qualified evaluator scored it against ISO 5060, and the quality record documents the verdict with the human accountability ISO 18587 requires. The chart is not a diagram of who does what. It is a diagram of who is accountable for what, and it is drawn so that when the deposition comes, and in regulated content it eventually comes, every question has an answer that is a person and a record rather than a shrug and an engine.

A good AI-enabled org chart is one that can answer the general counsel's questions out loud: who decided this was safe, and who signed off that it passed. If the honest answer to either is "nobody" or "the engine," the chart is not conformant and the deposition is only a matter of time.

Key Takeaways

• An org chart is a bet about where the hard work lives. When the engine took over first-draft production (MTPE at roughly 46% adoption, offered by 81% of LSPs), the scarce work moved from producing words to owning whether output is correct, terminology-faithful, and safe. A chart still optimized for volume is spending its senior authority on the one thing the machine already does.
• The Head of Localization AI is the single accountable executive owner of the AI-enabled operating model, engine and vendor strategy, risk-tiering policy, and P&L. It is defined by one power the old localization manager lacked: the standing authority to refuse to ship unsafe content without escalating past its own seat. Place it where that authority is real, reporting to enterprise risk or product quality, not marketing throughput.
• The Quality Lead owns the evaluation program, the MQM/ISO 5060 severity-scored delivery gate, evaluator qualification, and the audit-grade quality record. One Critical error fails the file regardless of an automatic QE score. The line must be structurally independent of production, because quality's job is to fail files and production's job is to ship them; if they share a boss whose bonus is on-time delivery, the gate becomes theater.
• The Terminology Lead owns the termbases as governed enterprise assets and the policy that makes approved terms hold through MT and post-editing. It needs its own seat because the engine's habit of drifting off an approved term to a fluent synonym is invisible, compounds through the translation memory, and can be a Critical error in regulated content. Under-ranking this role is a common and expensive mistake.
• The reorganization is mostly a re-pointing of authority, not a hiring spree: from "words per month" to "quality risk owned," from "reviser as a production step" to "an independent gate," from "terminology as a part-time favor" to "a governed asset owner." The pyramid inverts its emphasis so scarce senior authority and compensation sit on the quality-ownership roles.
• Grow these roles from within first. They are layerings on deep linguistic and domain competence, the hardest part to acquire, and growing them solves the retention problem by giving departing linguists a destination that moves them up rather than out. Avoid the anti-pattern of hiring a generalist "AI leader" with no localization depth; the revised ISO 18587's full-competence requirement codifies that you cannot own quality on machine output without the competence to judge it.
• The worked chart puts the Head of Localization AI at the top with Quality, Terminology, and Production reporting in as peers, so quality and terminology are never subordinate to the schedule. Three authority lines make it work: the independent gate line, the terminology grounding line, and the intake escalation line for potentially MT-forbidden content.
• The test of any AI-enabled org chart is whether it can answer the general counsel's two questions out loud: who decided this content was safe to machine-translate, and who signed off that it passed. If the honest answer to either is "nobody" or "the engine," the chart is not conformant under ISO 18587 and 5060, and the deposition is only a matter of time.