Nina Delgado spent two years attending working group meetings at the National Institute of Standards and Technology before she understood what she was actually doing there. She is a senior technology policy advisor at a Cabinet-level federal agency, assigned to represent the agency in the AI Risk Management Framework development process beginning in 2021. For the first eighteen months she showed up, listened, submitted written comments and reported back. She did not speak in sessions. She did not propose language. She did not build relationships with the private sector participants who were doing most of the active drafting. In year three the framework included a set of documentation expectations for federal AI systems that created significant compliance burden for her agency's legacy system portfolio, expectations she could have shaped differently had she understood earlier that showing up is not the same as participating. This lesson is about the difference between attendance and influence, and why that difference matters for agencies that deploy AI at scale.

Why government participation matters

AI technical standards are not abstract documents. They become operational requirements by a route that has nothing to do with whether they are legally binding. When the National Institute of Standards and Technology publishes AI guidance, federal agencies face pressure to align their governance practices with it even where the guidance is voluntary. When the International Organization for Standardization publishes a standard for AI system transparency, vendors treat conformance as a market differentiator and agencies begin referencing it in procurement. When the Institute of Electrical and Electronics Engineers develops standards for bias testing, those documents drift into being the working definition of due diligence in source selection and in inspector general reviews.

Agencies that do not participate are governed by standards shaped mainly by technology vendors, academic researchers and civil society organizations. All of those bring legitimate perspectives, and none of them operate under the constraints a state benefits agency, a federal permitting office or a municipal transportation department operates under. The specific realities of government AI, including legacy system integration, procurement cycle timelines, constitutional due process requirements, equity obligations, and transparency obligations under public records law, are not self-evident to a working group composed largely of private sector technologists. They become evident when a government participant makes them explicit, in the room, with text.

One clarification is worth making early, because it prevents a common misreading. Participating in the development of a voluntary framework does not make that framework binding on your agency, and adopting it does not discharge whatever statutory obligations already apply. A voluntary standard is a way of structuring good practice and of signalling seriousness, not a substitute for the rules your counsel already tracks. Agencies that treat conformance with a voluntary framework as legal cover have taken a genuinely useful instrument and asked it to do something it cannot do.

How standards actually get made

Understanding the process is the prerequisite for influencing it. Bodies differ, but most follow a similar arc. A working group or technical committee is convened. An initial draft is circulated for comment. Public comment periods typically run for 60 to 90 days. Comments are adjudicated by the working group. Revised drafts circulate. A final standard is published after some form of balloting. Each of those steps has a different influence profile, and most government participants engage at the step where influence is cheapest to attempt and least likely to work.

The critical insight is that the draft stage is where influence is most available and least exercised by government. By the time a document is in its final comment period the major structural decisions have been made. Organizations engaged since the working group convened have already shaped the draft's structure, terminology and scope. Late comments can correct errors and add nuance, but they cannot change the architecture of a document that has been built over 18 months by people who were present throughout. Structure is decided early and quietly, usually by whoever volunteered to write the first outline.

Nina's error was waiting until she understood the standard before trying to influence it. The people who shaped it most were learning alongside her, but they were also drafting alongside her, so their learning entered the text in real time while hers entered a notebook. Her understanding shaped nothing until year three, when she finally proposed specific language and discovered that the document's architecture made her agency's primary concern structurally awkward to address without reopening debates the group considered closed. Reopening a closed debate is possible. It costs credibility that a newer participant does not yet have.

Entry points: NIST, ISO and IEEE

Each of the three major bodies has a different structure for government participation, and the practical first step differs in each case.

NIST. Participation from federal agencies is invited as a matter of course, and formal mechanisms exist for interagency coordination on standards positions. Non-federal government bodies, including states, counties and municipalities, can participate through public comment and in some cases through formal liaison relationships. A participating organizations list is published for each major framework development, and getting onto that list is the first step toward active involvement. The source for this lesson states that the AI Risk Management Framework development included 240 organizations in active participation roles, and that less than 10 percent of them represented state or local government. Treat both figures as the source's account and verify the current participation record before quoting either in a briefing.

ISO. The organization works through national standards bodies. In the United States the member body is the American National Standards Institute, and agencies participate in ISO AI standards work through accredited Technical Advisory Groups. Membership in such a group requires an application and an annual fee, which the source for this lesson puts at $2,000 to $8,000 depending on the organization's size and the group's structure. Fee schedules change, so confirm the current figure with the body itself rather than budgeting from this lesson. For a federal agency or a large state, the order of magnitude is the point: this is a small line item for access to the process that produces internationally recognized AI governance standards.

IEEE. Standards development is open to individual and organizational members. Government employees can participate as individual members and contribute to working groups, and the body itself does not require agency authorization to do so. That is a fact about the standards body, not a clearance from your agency. Your own ethics rules, outside activity rules, and rules about who may state an agency position still apply, and the moment a working group treats you as speaking for your agency you are in a different category than an individual member. Clear the arrangement with ethics counsel before the first session and be explicit in the room about which capacity you are in. Agencies with real interests in a specific standard typically designate a representative anyway and allocate work time for it.

Laid side by side, the three routes differ less in difficulty than in what they ask of the agency. One asks you to be on a list and turn up. One asks you to join a national group and pay for the seat. One asks an individual to hold membership while the agency decides how much of itself stands behind that person. The table below summarizes the practical first step and the thing most likely to stop you.

Two cautions apply across all three. Structures and fee schedules change, so confirm the current arrangement with the body rather than relying on a training document. And confirm the current status and name of any organization before you cite it in a briefing, because bodies in this space are periodically renamed, merged or wound down, and a citation to something that no longer exists undermines everything around it.

Assessing readiness before you commit

Standards participation fails more often from bad setup than from bad performance in the room. Before committing, assess four things. Current capability: who in the agency has the subject matter depth to contribute, what gaps exist, and where the greatest opportunity to shape an outcome sits. Organizational readiness: is the agency prepared to have a position, and to be seen holding it in public. Stakeholder alignment: who inside the agency cares about this standard, what are their interests and concerns, and how aligned are they on direction. Resource constraints: what time, travel and fee budget genuinely exists, and how do you work inside that rather than assuming it will grow.

The alignment question is the sharpest. A standards representative with no internal mandate cannot propose language, because any text they offer might contradict a position another office holds. That is the mechanism behind Nina's two silent years more than any personal reticence: she had been sent to attend, not to commit the agency to anything, and nobody had told her which of the two she was doing. Settle that before the first session. The output of the alignment work is a short written statement of what the agency wants from the standard and what it cannot accept, approved by someone who can approve it.

Strategy, goals and risk

Treat participation as an initiative with a plan rather than a standing meeting invitation. Goal clarity comes first: what do you want the standard to say or not say, why does it matter to this agency in operational terms, and what would success look like when the final document publishes. Action planning follows: which sessions, which subgroups, which comment deadlines, in what sequence, with what preparation for each. Risk management is next: what could go wrong, how would you mitigate it, and what is the contingency if the draft goes somewhere you cannot accept. Stakeholder engagement runs through all of it, inside the agency as much as outside.

The risks worth naming are specific. Your representative may be reassigned mid-cycle, which resets relationships to zero. The agency position may change under a new administration while the draft is mid-flight, leaving your representative defending text the agency no longer wants. The standard may be adopted by reference into a procurement requirement faster than your systems can conform, which turns a document you helped write into a compliance problem you helped create. And a proposal you win early can be revised later by people who were not in that debate. Track the document to publication rather than declaring victory at adoption of your paragraph.

Building influence

Influence is built through three practices: consistent presence, specific expertise and credible proposals. None of them require seniority and all of them require time.

Consistent presence means attending working group sessions rather than only submitting written comments. Written comments are read and adjudicated, but the relationships and credibility that determine how a comment is received are built in sessions. Participants who attend consistently become familiar, are taken more seriously, and are far more likely to be asked onto the drafting subgroups where the most substantive influence is exercised. Presence buys standing rather than outcomes, which is worth saying plainly: you can attend every session for three years and still lose the argument. What you cannot do is win it from outside the room.

Specific expertise means contributing the knowledge government has and private sector participants typically lack. The operational reality of deployment inside procurement rules and legacy systems. The legal and constitutional requirements that shape government AI governance. The equity obligations that shape how public sector AI must be designed and evaluated. The transparency duties that come with public records law. This is distinctive value and working groups generally want it, when it arrives as substance rather than as an assertion that government is different.

Credible proposals means arriving with drafted language rather than concerns. "This requirement will be impossible for agencies with legacy systems to implement as written" is a concern. "Here is alternative language that achieves the same transparency goal without presupposing a modernized data architecture" is a contribution. The second is far more likely to reach the final text, because a working group under deadline adopts text that is already written more readily than it drafts a solution to a problem someone described. Bring the paragraph.

Coalitions and partnership

A single agency is one voice among many. Coalitions change that arithmetic, and the partners are not who government participants usually expect. Partnership design asks who the potential partners are, what each brings, and what the shared basis is: civil society organizations often want the same transparency provisions you want for different reasons, other agencies face the identical legacy constraint, and industry associations sometimes prefer a workable requirement to an aspirational one. Governance asks how a coalition decides a joint position, how disagreements get resolved, and who is authorized to speak for the group. Benefit sharing asks what each partner gets and whether the distribution holds up over a long drafting cycle.

Interagency coordination deserves separate attention because it is the cheapest coalition available and the most frequently skipped. Two agencies arriving with the same proposed language have materially more weight than two agencies arriving with similar concerns, and the coordination cost is a phone call before the session rather than a formal agreement. The failure mode is equally cheap to describe: two federal agencies submitting contradictory comments on the same clause, which resolves the question in favor of whoever else is in the room.

Running participation as a program

Good intentions do not survive a calendar. Capability building asks what skills the role requires, how they get built, and how they are sustained when the person holding them is promoted. Process design asks what repeatable procedures support participation: how a draft gets circulated internally for review, how comments are consolidated, how proposed language is cleared, and how a session is reported back so the knowledge lands somewhere other than one person's memory. Continuous improvement asks how you monitor whether participation is producing anything, how you spot what to change, and how the change actually gets made.

The internal clearance process is where most agencies lose their speed advantage. A working group moves on a drafting timescale of days, and an agency that needs three weeks to clear a paragraph will always be commenting on a decision instead of shaping it. Pre-clear a scope of positions rather than clearing each sentence: agree in advance the boundaries within which your representative may propose text without returning for approval, and reserve full clearance for positions outside that envelope. That single change converts an observer into a participant more reliably than anything else in this lesson.

Sustaining participation over a long cycle

Standards work is long. The source for this lesson describes the commitment as 4 to 8 hours per month of sustained engagement over 18 to 36 months. Multiplying those stated inputs gives a range of roughly 72 to 288 hours across the cycle, and that arithmetic is worth doing in front of the executive who will approve the time, because the monthly figure sounds trivial and the total does not. Ad hoc participation by staff carrying a full program workload does not produce influence; a designated function with explicit time allocation does. Funding sustainability asks how the fees and travel are paid and what happens when that source changes. Organizational embedding asks whether participation is a role or a person.

Succession is the failure that ends most government participation. The relationships, the history of which arguments were had and settled, and the sense of who will support which position all live in one person's head, and when they move the agency starts from zero in a process where newcomers have no standing. The countermeasure is unglamorous: a running file of positions taken, language proposed, outcomes, and the names and interests of the other active participants, maintained as the work happens rather than assembled at departure. Send a second person to sessions occasionally, even without a speaking role, so the institutional memory has somewhere else to live.

What changed for Nina

After year two Nina began drafting proposed language before sessions rather than arriving with observations. By year four she had proposed three specific language modifications that were adopted in the final framework, two of them substantially unchanged. None of them required expertise she did not already have when she was sitting silently in year one. All of them required translating that expertise into the currency of standards development, which is concrete, specific, defensible text that a working group under deadline can drop into a draft.

The second change mattered as much and is easier to copy. She secured an internal envelope of pre-cleared positions, so she could offer text in the room without a three-week round trip, and she started sending a colleague to alternate sessions so that the agency's presence did not depend entirely on her calendar. Neither change required a budget increase. Both changed what her attendance was capable of producing.

Anti-patterns

• Attendance mistaken for participation. Showing up, listening and reporting back for months while the drafting subgroups shape the architecture. Presence is the entry condition, not the contribution.
• Comments without text. Submitting concerns and expecting the working group to invent a solution. Concerns get adjudicated; drafted language gets adopted.
• Arriving at the final comment period. Engaging when the document is nearly finished, when the only available changes are the ones that do not disturb a structure decided long ago.
• The individual member loophole. Treating a standards body's willingness to accept individual members as clearance from your own agency. Ethics rules, outside activity rules and rules about stating agency positions still apply, and the room will hear your agency's name whether or not you invoke it.
• Voluntary framework as legal cover. Treating conformance with a voluntary standard as discharge of a statutory obligation. It is evidence of diligence and it is not a defense, and a framework you helped write is not thereby binding on anyone.
• No internal mandate. Sending a representative with no agreed agency position, so that every possible proposal risks contradicting some other office and the safe move is silence.
• Clearance slower than the draft. Requiring full internal approval for each sentence in a process that moves in days, which guarantees that the agency always arrives after the decision.
• Contradictory federal voices. Two agencies filing opposing comments on the same clause, which hands the decision to whoever else is in the room and costs both of them credibility.
• Victory declared at adoption. Winning a paragraph early and stopping attendance, so the text is revised later by participants who were not part of the original debate.
• Single point of memory. Letting the relationships, history and positions live in one person's head, so a reassignment resets the agency to the standing of a newcomer.

Practice prompts

1. Pick one AI standard or framework that already shapes requirements in your agency. Write down what it requires, who wrote it, and whether anyone from your agency was in the room. Note where you had to guess.
2. For standards participation in your agency, run the readiness assessment: current capability, gaps, organizational readiness, stakeholder alignment and resource constraints, including fee and travel budget.
3. Draft the one-page internal mandate you would need before speaking in a session: what the agency wants from the standard, what it cannot accept, and who approved that.
4. Take one operational constraint your agency lives with, such as a legacy integration limit or a records obligation, and write it as proposed standard text rather than as a concern.
5. Design the participation program: which capabilities to build, what internal circulation and clearance process to run, how sessions get reported, and what improvement mechanism keeps it honest.
6. Identify two potential coalition partners for a position you hold, including at least one outside government, and write the shared sentence you could both sign.
7. Design the measure of whether participation is working: what you will count, who collects it, who reads it, and what result would justify stopping.

Reflection

Nina's two silent years were not wasted, but they were expensive, and the expense was invisible until a documentation requirement landed on a legacy portfolio that nobody in the drafting room had ever had to operate. That is how standards costs usually arrive in government: not as a rule imposed by an unfamiliar body, but as a reasonable requirement written by reasonable people who had never seen the system it would apply to. Think about the standards and frameworks currently shaping your own agency's AI work. Ask which of them contain a requirement that would have been written differently if someone from your side of the work had been present with a paragraph. Then ask what it would actually take to be that person, and whether the barrier is authority, time or simply that nobody has asked.

Glossary

• Standards development organization. A body that convenes participants to write technical standards through a defined process of drafting, comment, adjudication and balloting.
• Working group. The group convened to draft a specific standard, within which most substantive text decisions are made.
• Drafting subgroup. A smaller set of participants who write particular sections. Membership is usually by invitation and is where influence concentrates.
• Public comment period. The defined window in which anyone may submit comments on a draft, after which comments are adjudicated by the working group.
• Adjudication. The process by which a working group decides what to do with each submitted comment, including accepting, modifying or declining it with a reason.
• Technical Advisory Group. An accredited national group through which participants contribute to an international standards committee.
• National standards body. The organization that represents a country in an international standards organization and channels national participation into it.
• Voluntary standard. A standard that carries no legal force by itself, but which can acquire practical force through procurement references, audit expectations and market adoption.
• Adoption by reference. The practice of a contract, regulation or policy requiring conformance with an external standard, which converts a voluntary document into a binding requirement for the parties.
• Internal mandate. The written, approved statement of what a representative may say and propose on the agency's behalf without returning for clearance.
• Governance. The structure and process by which decisions are made and disagreements resolved, whether inside an agency or across a coalition.

Related lessons

• Contributing to Standards Bodies (NIST, ISO, IEEE, OECD) covers the mechanics of contribution across bodies in more detail.
• ISO 42001: AI Management System Design covers what conformance with a management system standard actually asks of an agency.
• NIST AI RMF: The GOVERN Function covers the framework Nina was working on, from the implementation side.
• International Standards: EU AI Act and OECD covers the international instruments that interact with technical standards work.
• International Government AI Collaboration covers the wider diplomatic context in which standards positions get coordinated.
• Cross-Agency AI Coordination covers the interagency mechanics behind a consistent federal voice.
• Leading Government-Wide AI Initiatives covers building the internal mandate that standards participation depends on.
• Public Consultation on AI Policy covers the comment process from the side of the body running it.

Closing

Standards work rewards a kind of patience that government is actually good at when it decides to be. The barrier is rarely expertise, because agencies hold operational knowledge that drafting rooms genuinely lack. The barrier is that participation looks like a discretionary meeting until the day a requirement lands, and by then the cheap moment has passed. Send someone, give them a mandate and an envelope of pre-cleared positions, protect the hours, keep a written record so the next person inherits standing rather than a calendar invitation, and require them to bring text. Everything else in this lesson is detail attached to those five decisions.

Key takeaways

• Attendance is not participation. The draft stage is when standards are shaped. Organizations engaged from the convening onward have structural influence that late commenters cannot obtain.
• Government holds expertise the room lacks. Procurement constraints, legacy realities, constitutional requirements, equity obligations and public records duties are not self-evident to private sector participants. Make them explicit, early, as text.
• Each body has a different entry point. A participating organizations list and public workshops at NIST; an accredited Technical Advisory Group for ISO work, reached through the national member body; individual or organizational membership for IEEE.
• The standards body's rules are not your agency's rules. Individual membership does not displace your ethics, outside activity and agency position rules. Clear the arrangement before the first session.
• Draft language before sessions, not after. A working group under deadline adopts text that is already written far more readily than it drafts a solution to a described problem.
• Get an internal mandate and a pre-cleared envelope. A representative who must clear each sentence will always arrive after the decision. Agree the boundaries in advance and reserve full clearance for what falls outside them.
• Build coalitions, starting with other agencies. Two organizations proposing identical language carry more weight than two raising similar concerns, and contradictory federal comments hand the decision to someone else.
• A voluntary standard is not legal cover. Helping write a framework does not make it binding, and conforming to it does not discharge statutory obligations that already apply.
• Protect the hours and write the range down. Sustained engagement of 4 to 8 hours a month over 18 to 36 months is what the source describes, and the total is what an executive needs to see before approving it.
• Plan for succession from the first session. Keep a running file of positions, proposed language, outcomes and other participants, and send a second person occasionally, so a reassignment does not reset the agency to zero standing.

Frequently Asked Questions

We are a state or local agency. Is standards participation realistic for us?

It is, and the underrepresentation is itself the argument. The source for this lesson reports that less than 10 percent of participants in one major framework development represented state or local government, which means the standards that end up governing state and local AI are largely written without state and local input. Public comment costs nothing but time. Liaison arrangements exist in some processes. The realistic constraint is sustained hours rather than eligibility, and one well-prepared comment with proposed text outperforms three years of silent attendance.

How do I know which standard to engage with when several are in development?

Work backward from your own systems. Identify the AI systems whose operation would be most disrupted by a new documentation, testing or transparency requirement, then find out which drafts touch that ground. Engaging with one document properly beats sampling four. If you genuinely cannot choose, ask which draft is most likely to be adopted by reference into a procurement requirement, because that is the route by which a voluntary document becomes something your contracting officer enforces.

Can I propose language that reflects my agency's position without clearing it first?

Not safely, and this is worth settling before it comes up in a session. The workable arrangement is a pre-cleared envelope: an agreed scope within which you may offer text without returning for approval, with anything outside it reserved for full clearance. Without that, either you clear every sentence and arrive too late, or you improvise and eventually commit the agency to something it did not agree to. Both outcomes are avoidable with a one-page document approved before the cycle starts.

Our representative is leaving. How do we avoid starting over?

Overlap them if you possibly can, and in the meantime capture what is not written down: positions taken and why, language proposed and its fate, which other participants are reliable allies on which issues, and what has already been argued and settled. Standing in these processes is personal, so some loss is unavoidable, but the difference between a successor who inherits that file and one who inherits a meeting series is roughly a year of effectiveness.

Is it worth participating if the standard will be voluntary anyway?

Usually yes, because the voluntary status is about legal force rather than practical force. Voluntary standards enter procurement documents, audit expectations, vendor marketing and inspector general questions, and at that point they shape your work whether or not anyone ever made them binding. The corollary matters too: because they are voluntary, conforming to one does not discharge a statutory obligation, so participate to shape the practical requirement without treating the result as legal protection.

What does a first year of participation actually look like if we do it well?

Mostly reading, one written mandate, and text. Join the relevant list or group, read the existing drafts and the comment adjudications from prior cycles, get the one-page internal mandate approved, attend consistently enough to be recognized, and submit at least one comment that contains drafted language rather than a concern. That is a modest year and it puts you ahead of where Nina was at the end of her second. Volunteering to write a section outline, if the opportunity appears, moves you further than anything else on this list.