Elena Vasquez, the chief AI officer of a large state government, spent two years and roughly $3 million building one of the country's better frameworks for procuring trustworthy AI. It worked. Then she watched a new international standard get published that contradicted three of her core requirements, and her vendors started citing it back to her. "We comply with the ISO standard," they said, "so your extra clauses are non-standard burden." She had built something excellent inside her own walls while the rules that would actually govern her market were being written, in rooms she was not in, by people who had never run a benefits system. The lesson landed hard: in AI, if you are not at the standards table, you are on the menu.

This lesson is for agency heads and national AI leaders who have built enough to know that the next frontier of influence is not internal policy but the external standards that shape everyone's policy. Contributing to standards bodies is how one agency's hard-won operating experience becomes the default that thousands of agencies and vendors inherit. It is the highest-leverage and slowest-paying work in this field, and most government leaders never attempt it. The mechanics of running participation as an ongoing program are covered in AI Standards Development Participation; this lesson is about the strategic choice of where to spend your voice and what your government uniquely has to say.

Why Standards Are Where the Real Power Sits

A standard is a written agreement on how something should be done, produced by a recognized body and adopted voluntarily across an industry. It sounds dry. It is one of the most consequential forms of soft power in technology, because standards become the floor that procurement contracts, regulations, and audits all build on. When a standard says trustworthy AI requires a particular practice, vendors build that practice, regulators reference it, and any agency that wants something different has to justify the deviation rather than simply asking for it.

Elena learned this backwards. The standard her vendors invoked became, overnight, the definition of reasonable in her own negotiations. She had not lost an argument about her clauses. She had lost the prior argument about what counts as normal, and that argument had been settled in a committee room two years earlier. Whoever writes the standard writes the default, and defaults govern far more behavior than rules do, because most people never push past them.

Government has a profound interest in those defaults, and a weak record of defending it. Standards work is expensive, slow, and invisible, which means it is dominated by the organizations that can afford to fund people to sit in rooms for years. Those organizations are overwhelmingly vendors. Their incentives are legitimate and they are not the public's. A standard written mostly by suppliers will be technically excellent and will quietly encode what is convenient to sell.

The Four Bodies and What Each Is For

The four named in this work are not interchangeable. Each has a different character, a different speed, and a different point of entry. Pick by what you are trying to influence rather than by prestige, because the body that matters for your problem is often the least glamorous one.

• NIST, the U.S. National Institute of Standards and Technology, produces influential frameworks and guidance, most notably the AI Risk Management Framework. That framework is voluntary and non-binding; it carries weight because agencies and buyers choose to reference it, not because it is law. NIST works largely through open public workshops and comment periods, which makes it the lowest barrier to entry and often the highest relevance for a U.S. agency.
• ISO, the International Organization for Standardization, working with the IEC on electrotechnical matters, produces the formal international standards that vendors cite globally, including the AI management system standard ISO/IEC 42001. Influence here runs through your national standards body and the delegation it sends. It is slow, formal, and consequential, because these are the documents international suppliers organize their compliance programs around.
• IEEE, the Institute of Electrical and Electronics Engineers, produces technical and ethical standards, including its work on ethically aligned design. It is more engineer-driven and academically inflected than the others, and it is comparatively open to individual technical experts rather than only to national delegations.
• OECD, the Organisation for Economic Co-operation and Development, sets high-level principles among governments, including the OECD AI Principles that many national policies trace back to. These are non-binding recommendations rather than rules. Influence here is government to government, and it shapes the values layer that sits above the technical standards.

For Elena, the fastest correction was NIST's open processes, where her state's procurement experience could shape U.S. guidance directly and within a comment window rather than a decade. The longer game was persuading her national standards body to carry her concerns into the international committee that had blindsided her. Those two moves have completely different costs, and confusing them is how agencies end up doing neither.

What Conformance Does and Does Not Prove

Elena's vendors made a claim worth taking apart carefully, because a version of it will be made to you. Conformance to a management system standard describes how an organization runs its processes: that it has assigned responsibilities, documented its decisions, assessed risks, and set up review. That is genuinely valuable and it is not the same as evidence that a particular model is accurate, fair, or lawful in your particular use. A supplier with an excellent management system can still ship a model that fails badly on your population.

Nor does conformance displace your obligations. A voluntary standard is not law in itself. It acquires force when a contract requires it, a regulator references it, or a market treats it as the price of entry. That means a vendor asserting that a standard makes your additional clauses illegitimate is making a commercial argument dressed as a technical one. Your clauses may be poorly drafted or unnecessary, and that is a fair conversation, but the standard does not settle it. The honest formulation is that a standard raises the floor and your accountability sits above the floor.

This matters for how you argue inside a standards process too. The most persuasive government contribution is rarely a demand for more requirements. It is a demonstration that a proposed requirement will not survive contact with a real public service, or that a claimed control does not do what the drafters believe it does.

How a Standard Actually Gets Written

Standards feel like they descend from on high. They do not. They are written by working groups of people who show up, draft text, argue over single words, and outlast each other. Understanding the rough lifecycle tells you where a government voice can intervene and what each point of entry costs.

1. Proposal. Someone argues that a standard is needed. Influence at this stage shapes whether the document exists at all and, more importantly, what it covers and what it deliberately leaves out.
2. Working group. Volunteers draft the text over months or years. This is where the substance is decided and where consistent presence wins, because the person in the room when a clause is redrafted is the person whose concern gets accommodated.
3. Public comment and balloting. Drafts go out for review. This is the lowest-effort entry point, where a well-argued and evidenced comment can still change specific clauses.
4. Publication and revision. The standard ships, then gets revised on a cycle. The work never truly ends, which means a missed window comes around again and a clause you won can also be unwound later.

The practical implication for a busy agency head is a matching problem. The highest-influence stage demands sustained presence over years, while the comment stage offers real influence for a few days of sharp effort. Match your ambition to the time you can actually commit, start at the comment stage, and let a demonstrated track record of useful comments earn you the invitation into the drafting work.

What Government Uniquely Brings to the Table

Government's seat in standards work is underused, and that is a problem, because government brings something no vendor can: the perspective of the public, and of operating real systems at scale under legal accountability. A vendor can describe how a control is supposed to behave. An agency can describe what happened when it was applied at volume, by caseworkers under time pressure, to determinations that changed people's incomes.

Your unique contributions are concrete rather than philosophical. You bring failure data from running systems that affect millions of people, which almost no other participant has and which is difficult to argue with. You bring the accountability lens, meaning what actually happens when a model contributes to denying someone a benefit and they appeal, complain, or sue. You bring the equity perspective that other participants tend to treat as an optional module. Elena's lived experience, that a particular transparency requirement was unworkable in a high-volume benefits context, was exactly the input the drafters had lacked. Her absence was the reason the published text contradicted reality.

There is a discipline that comes with this. Government evidence carries weight because it is specific, so bring the case, the volume, the failure mode, and the cost, not a position paper about values. A committee will move for a concrete account of why a clause breaks. It will not move for an assertion that the public interest requires something.

An Influence Strategy You Can Actually Execute

You cannot personally sit in every working group, and pretending otherwise produces the worst outcome, which is thin presence everywhere and influence nowhere. The strategy is to be deliberate about spending a genuinely scarce resource: credible expert time. Three principles separate participation that changes text from participation that fills a seat.

• Pick your battles. Choose the one or two standards that most affect your mission and go deep there rather than thin everywhere. Depth is what converts attendance into drafted language.
• Send sustained representation rather than symbolic representation. A senior person who appears once accomplishes less than a capable expert with a protected mandate to participate for years. Fund the seat as a real part of someone's job with time written into their objectives, not as a volunteer favor absorbed on top of a full workload.
• Work as a coalition, not solo. Other agencies and other levels of government share your concerns and rarely know it. A coordinated position carried by several bodies, or through an intergovernmental association, is far harder to characterize as one jurisdiction's idiosyncrasy.

Add one unglamorous fourth practice: write the position down before you go. A short paper stating the specific change you want, the evidence for it, and the fallback you would accept turns a delegate into a negotiator. Without it, your representative will be sympathetic, well liked, and unable to explain what would satisfy them, which is how good intentions produce no change to the text.

Standards Engagement Plan

Use this to decide, concretely, where your agency will engage and at what depth. Filling it in forces the choices that turn good intentions into actual influence, and it makes the resourcing question visible before you commit someone's name to a committee. The worked column reflects Elena's corrected strategy.

Bringing the Win Home

The row in that plan most agencies leave blank is the feedback loop, and it is the row that determines whether any of this was worth doing. Influence spent outward has to return inward or it stays a hobby. When Elena's language was adopted, the value was not the citation. It was that her state's procurement templates could then require the practice by pointing at a document her vendors already claimed to follow, which turned a contested clause into an administrative formality overnight.

Design that return path before you start, because it is what makes the work legible to people who will never read a standard. Decide in advance which of your own instruments will absorb an adopted clause: procurement templates, agency policy, review criteria, audit checklists, or the questions your board asks before approving a deployment. Name the office that owns each one and tell them what is coming, so the update happens on publication rather than eighteen months later when someone notices.

The return also runs the other way, and this is the part leaders underestimate. Participating in a drafting process gives you sight of requirements a year or more before they reach your market. That advance warning is worth funding on its own terms even in the years when your language does not survive the committee, because you get to prepare your suppliers, your contracts, and your staff while everyone else is still surprised. Report that benefit explicitly, because it arrives on a schedule your leadership can recognize.

The Payoff Is Slow, and It Is Not Automatic

Standards work pays nothing this quarter. Its return is measured in years and in scale, which makes it structurally hard to fund inside institutions that report quarterly. The transparency requirement Elena could not change after publication she might have shaped before it, with a few days of well-argued comment two years earlier. That is the whole economics of this work: the cheapest moment to influence a standard is always before anyone cares about it.

Be precise about the size of the win, though, because overselling it is how these programs lose their funding at the first review. A clause you get into an international standard governs the organizations that adopt that standard, for as long as that edition stands. Adoption is voluntary and uneven, revisions can unwind your language, and a competing document can supersede the whole thing. What you have bought is a strong and durable default, not a permanent rule. That is still more reach than any internal policy you will ever write.

Elena did not undo the standard that blindsided her. She funded a real seat for the next revision, joined a coalition of peer states, and submitted detailed comments to NIST that shaped its next profile. Three years on, the revised international standard carried language she had drafted about accountability in high-volume public-sector decisions. Her vendors now cite that clause back to her, and it says what the public needed it to say. She had moved from the menu to the table, and the only thing that had changed was that she showed up.

Anti-Patterns to Avoid

• Treating a vendor's conformance claim as a passed test. Certification against a management system standard describes process discipline, not the behavior of a specific model on your population, and it never displaces your own legal obligations. Ask what was assessed, by whom, and against which scope.
• The prestige seat. Sending the most senior available person to the most famous available committee, once. Influence in standards accrues to whoever is present when the text is redrafted, which is rarely the person with the largest title.
• Arriving with values instead of evidence. Committees move for a specific account of why a clause fails in operation. A statement that the public interest requires something is easy to agree with and easy to ignore.
• Volunteering the seat on top of a full job. Unfunded participation collapses at the first operational crisis, which is exactly when the drafting gets interesting. An unfilled seat is worse than never claiming one, because you stop watching.
• Chasing every relevant standard. Thin presence across many committees produces a report full of engagements and no changed language. Two deep commitments beat a long list of shallow ones.
• Selling the win as permanent. Describing an adopted clause as a rule that now binds the field. Standards are voluntary, adoption is uneven, and the next revision can undo your text; overstating the result is how the next budget cycle kills the program.
• Confusing the fast door with the slow door. Assuming an open comment process and a formal international committee take comparable effort. Budgeting for one and expecting results from the other is how agencies conclude that standards work does not pay.
• Waiting to be invited. Public comment periods are open, published, and routinely under-subscribed by exactly the operators whose experience the drafters most need. Nobody will ask you.

Practice Prompts

• Find the standard already governing you. Pick one live AI contract and identify every external standard, framework, or profile it references. Establish who wrote each and whether anyone from your level of government was in the room.
• Draft one comment. Take an open consultation or draft document relevant to your mission and write a comment that names the specific clause, the operational failure it would cause, the evidence from your own systems, and the alternative wording you would accept.
• Cost the seat honestly. For the one standard you care most about, work out what sustained participation would actually require in a named person's time over a multi-year cycle, and decide whether you will fund it or drop the ambition.
• Assemble the evidence packet. Collect the failure data, volumes, and appeal outcomes from your own systems that no vendor participant could produce. That packet, not your title, is your negotiating position.
• Find your coalition. Identify three peer bodies likely to share your concern on one specific clause and test whether a joint position is possible. Note where their interests diverge from yours, because that is where the coalition will break.

Reflection

Think about the external documents that currently define what counts as reasonable in your procurement, your audits, and your policy arguments. Ask who wrote each of them, and whether anyone with your operating experience was present while they were drafted. Then ask the more uncomfortable question: how many open comment opportunities relevant to your mission have passed without a submission from your organization, and what would it actually have cost to answer one of them well. The gap between how much those documents govern you and how little effort you spend shaping them is the honest measure of this lesson.

Glossary

• Standard. A written agreement on how something should be done, produced by a recognized body and adopted voluntarily, which acquires practical force when contracts, regulators, or markets require it.
• Voluntary framework. Guidance an organization chooses to adopt, such as the NIST AI Risk Management Framework, which is influential and non-binding rather than legally mandatory.
• Management system standard. A standard describing how an organization governs a domain, covering responsibilities, documentation, risk assessment, and review, as distinct from a standard specifying how a product must perform.
• Conformance. A claim that an organization or product meets a standard's requirements. Its value depends entirely on what was assessed, by whom, and against what scope.
• Working group. The committee of participants who draft and redraft a standard's text over months or years, and the point in the process where substance is actually decided.
• Public comment period. A published window during which anyone may submit views on a draft document, and the lowest-cost point of entry for an under-resourced participant.
• National standards body. The organization through which a country participates in international standards work and forms the delegation that carries national positions into committees.
• Non-binding recommendation. An agreed statement of principle among governments, such as the OECD AI Principles, that shapes national policy without creating a legal obligation in itself.
• Default. The option that applies when nobody makes an active choice, and the reason standards govern more behavior than the rules that reference them.

Related Lessons

• AI Standards Development Participation covers the operating mechanics of participation, including readiness assessment and running it as a sustained program.
• International Standards: EU AI Act and OECD examines the instruments themselves and what obligations they do and do not create.
• International Governance Coordination addresses how governments align positions before they reach a standards committee.
• International AI Diplomacy develops the government-to-government channel through which principles-level influence travels.
• Publishing on Government AI covers building the public record and credibility that makes your standards contributions carry weight.
• Navigating the Federal AI Landscape maps the domestic bodies and processes a U.S. agency engages before it looks abroad.
• Multi-Level Government AI Governance takes up how shared standards hold independent levels of government together.
• National AI Competitiveness situates standards influence within a nation's wider strategic position.

Closing

The uncomfortable truth in Elena's story is that nothing about her defeat was technical. She lost because a document was written without her, and she recovered because she arranged, at modest cost, to be present the next time one was. Standards work will never compete for attention with a launch, a crisis, or a budget hearing. It has no constituency inside your organization and produces nothing you can point to for years. It is also the only lever you have that keeps working after your policy is rewritten, your platform is replaced, and your successor's successor has taken the seat. Fund one person, pick one document, and start with a comment.

Key Takeaways

• Standards are the floor everything else builds on. Whoever writes the default governs more behavior than any single regulation, because most actors never push past a default.
• If you are not at the table, you are on the menu. Standards written mostly by suppliers will be technically excellent and will encode what is convenient to sell.
• Match the body to the goal. NIST is the most accessible route for U.S. agencies and its framework is voluntary, ISO carries the international documents vendors organize around, IEEE is engineer-driven, and OECD sets non-binding principles among governments.
• Conformance is process evidence, not a safety result. A supplier's certification says how it runs itself, not how its model behaves on your population, and it never displaces your own obligations.
• The working group decides substance. Public comment buys real influence for a few days of sharp effort; changing the drafting requires sustained presence over years.
• Bring evidence, not values. Failure data, volumes, and appeal outcomes from real public services are the contribution no other participant can make.
• Fund a real seat with a written position. A named expert with protected time, a specific change to request, and a stated fallback beats a senior visitor with good intentions.
• Build a coalition. A position carried by several bodies is far harder to dismiss as one jurisdiction's idiosyncrasy.
• Claim the win accurately. An adopted clause is a strong, durable default among those who adopt that edition, not a permanent rule, and overselling it is how the program loses its funding.

Frequently Asked Questions

We are a state or local body, not a national government. Can we influence international standards at all? Directly, rarely; indirectly, more than you would expect. International committees are reached through national delegations, so your route runs through your national standards body and through whichever federal counterparts already hold seats. What you have that they lack is operating evidence from delivering services to actual people, and delegations are usually short of exactly that. The more accessible path in the meantime is the open comment processes on domestic guidance, where your submission goes straight into the record.

Is it worth commenting if we cannot join the drafting? Yes, and it is the highest return per hour available in this work. Comment periods are routinely under-subscribed by operators, so a submission that names a specific clause, describes the operational failure it would cause, and proposes alternative wording stands out simply by being concrete. It also builds the track record that makes an invitation into the drafting work plausible later. A comment that says the draft should be more mindful of the public interest will change nothing.

How do we respond when a vendor says our contract clauses exceed the standard? Separate the two questions the vendor has merged. First, does the standard actually address the risk your clause addresses, and at what level of specificity? Frequently it does not, because management-system requirements and application-specific controls are different things. Second, is your clause well drafted and proportionate, which is a fair challenge and worth answering on the merits. What you should not accept is the implied premise that conformance to a voluntary standard defines the limit of what a public buyer may require.

How do we justify multi-year standards work to leadership that reports quarterly? Do not promise an outcome you cannot control, because you cannot guarantee that a committee adopts your language. Promise what participation reliably delivers: early sight of requirements before they land on your procurement, relationships with the people drafting them, and a documented position that makes your own policy easier to defend. Then report the intermediate milestones, a submitted comment, a seat obtained, a clause put in play, so the work has visible progress long before it has a result.