What it Takes to Standardize AI
If we want human rights and the public interest built into AI systems, we need to ensure that they are written into the technical standards that decide how these systems are designed and interoperate.
Civil society advocates are heading to Geneva next month for the UN's inaugural Global Dialogue on AI Governance (6–7 July) and the ITU AI for Good consultation (8 July). I prepared notes for a planning meeting with GNI's Multistakeholder Approaches to Participation in AI Governance (MAP-AI) initiative, a coalition working to elevate underrepresented voices across critical AI governance processes. These are high level reflections on what it will take to make that participation count, so watch this space for deeper analysis as 2026 shapes up to be a pivotal year for AI standards.
If we want human rights and the public interest built into AI systems, we need to ensure that they are written into the technical standards that decide how these systems are designed and interoperate. It's critical that values like privacy are not bolted on afterward, or worse, forgotten altogether. That means being in the room where those choices are made. Over the next year our hope is that many civil society advocates show up to build a stronger public-interest foundation for AI governance at the meetings and events where standards are written.
AI systems are core infrastructure for information access, economic opportunity, and civic participation. At the same time their development and deployment are increasingly shaped by opaque technical decisions, proprietary architectures, and fragmented governance processes. As Ada Lovelace Institute pointed out after an expert roundtable on EU AI standards development and civil society participation: “One of the key issues identified by participants was the lack of resources committed to ensuring experts in fundamental rights can engage meaningfully in the process.” This creates significant risks for human rights, democratic accountability, competition, and public trust. And mitigating this risk only gets more difficult over time since AI systems are multidimensional and deployed globally across very different legal jurisdictions and political contexts.
In order to ensure human rights considerations and governance are embedded in AI systems, it is necessary to strategically engage with appropriate depth and breadth in emerging AI standardization efforts. And to make that engagement, well, standard and routine.
Standards challenges are intensified
Years of work to consider human rights in internet standards have been met with consistent friction. These persistent problems, coupled with novel technical challenges, create a reinforced-concrete ceiling on effective public-interest engagement in technical governance of AI.
Public-interest participation in AI standards and technical governance is low. Seemingly narrow technical decisions, like how models are evaluated, how context and provenance are represented, how systems interoperate, or how accountability is surfaced, can have far-reaching societal consequences. Yet these decisions are overwhelmingly made by private actors within technical standards bodies and industry-led processes, often without sustained public-interest participation or human-rights expertise. Commercial organizations have the capacity to form well-funded entities like the Data & Trusted AI Alliance, a group of companies that include Meta and Walmart, to engage in standards like data provenance.
Alignment and understanding across technical, policy, and advocacy communities is remote. Many participants in standards processes often cannot hold the bigger picture in mind. Engineers lack familiarity with human rights, political economy, competition policy, and social impacts. AI governance is increasingly mediated through technocratic spaces where public-interest capacity is thin, coordination is weak, and incentives favor concentration over accountability. As regulation becomes inevitable, the absence of strong public-interest technical foundations risks locking in harmful design choices before democratic oversight can meaningfully intervene.
Technical capacity available to policymakers and advocates is inadequate. Public debates about AI governance, ranging from safety and transparency to competition and accountability, should be grounded in more technical claims, but often are not. Without trusted public-interest technologists able to assess, contest, and translate these claims, policy processes risk being captured by narrow technical narratives that privilege incumbent interests.
A fragmented and adversarial standards landscape is complicated by misaligned incentives. Unlike the internet, AI lacks strong built-in incentives for interoperability. In many cases, the incentive is the opposite: proprietary architectures, closed ecosystems, and strategic fragmentation. The lack of interoperability between agents is an acute example: its absence directly threatens established businesses like banks, yet almost no public-interest actors are engaged on it. It’s not that technical mechanisms are lacking, it is that adoption is not enforced in a technical or market-driven way. At the same time, there is no single institution mandated to exclusively govern AI. Instead, standards and norms are being set simultaneously across ISO/IEC, IEEE, ETSI, ITU, W3C, IETF, and industry-led initiatives, requiring public-interest actors to engage in many more venues with far fewer resources. Meanwhile, corporate and state actors are increasingly coordinated, well-resourced, and effective at shaping these processes to their advantage.
Better tech, better governance
Rather than working to change these acute problems for AI standards and governance, a smart strategy accepts them and works within them as constraints. The main constraints always being time and funding, civil society engagement can be effective and efficient with an incisive depth and breadth approach.
Work in depth on each phase of AI systems, from standards design to intermediaries' choices and management decisions regarding the ways human rights are impacted, rather than advocating for after-market patches to address harms. A focus on MCP, OAuth, C2PA and other emergent standards that are in use, particularly by government institutions, makes greater impact than a general monitoring function. The field moves fast, however. New standards can show up with little warning, the way MCP did, so it’s important to have an established presence in a standards body and then engage early in emergent work. Privacy is a key opportunity for engagement as AI developers aren’t focused on it but public interest advocates in internet standards are. The outcome is better technology.
Work broadly with scholars, implementers, policy makers, investors and civil society across all of these standards venues to translate real-world impacts into concrete standards requirements and deployment guidance, which leverages the human rights framework in order to set norms and provide a substitute for top-down policy cover. Setting agendas, enforcing mandates, and making connections across standards and governance bodies is a job civil society is particularly good at, given its geographic spread and distance from profit making. The outcome is better governance.
Tactically speaking advocates need to focus on the where, which dictates outputs, and the who, which differentiates the approach civil society has no choice but to take if human rights and the public interest are to be considered in the development and governance of AI. It also makes sense to organize engagement by sector (financial services, education or health, for example) where institutional buy-in is stronger and the path to regulatory impact may be clearer.
The standards landscape for AI is complex, fragmented, and moving quickly. It’s also not very durable either. Those are not problems that will be solved before meaningful public-interest engagement happens, these are things civil society is good at fixing.
The work ahead is therefore less about picking the winner for AI standards and more about building consistent and coordinated participation across many venues, connecting technical decisions to their social consequences and reaching consensus about tradeoffs. If human rights and the public interest are to shape AI systems, they must be valued as essential considerations within the technical conversations that build those systems.
Want to edit IX?
I'm taking a summer vacation! If you're interested in editing Internet Exchange for three weeks, from July 20–August 7, get in touch with your day rate, tell us a bit about your experience and share a few examples of your work. The time commitment is about 1.5 days per week.
The work involves copy editing, commissioning, structural edits, and working with writers. Fully remote. We're reviewing applications on a rolling basis, so get in touch soon. Email us at editor@exchangepoint.tech.
Support the Internet Exchange
If you find our emails useful, consider becoming a paid subscriber! You'll get access to our members-only Signal community where we share ideas, discuss upcoming topics, and exchange links. Paid subscribers can also leave comments on posts and enjoy a warm, fuzzy feeling.
Not ready for a long-term commitment? You can always leave us a tip.
This Week's Links
- More than 100 organizations and individuals, including employees of Google, OpenAI, and Anthropic, are calling on tech companies to stop providing AI systems for use in military kill chains, citing the role of AI targeting tools in US and Israeli strikes on Iran and Gaza and arguing that no technical or procedural fix can make such systems compatible with international humanitarian law. You can join them by signing on to their open letter:
