Innova Quantum Leap

Success for the first Global Dialogue on AI Governance should not be measured by the volume of commitments made in the room, but by whether the communities most harmed by algorithmic systems had genuine authorship over what was agreed. Three outcomes would make this Dialogue meaningful rather than ceremonial. First, the Dialogue must produce a named accountability mechanism for algorithmic distortions already in circulation, not only frameworks governing future AI development. Current governance instruments address consent and process rights for data use going forward. None provide a structured legal pathway for communities to contest and correct distortions that have already traveled through digital systems, been cited in policy, and shaped institutional decisions made about them. This Correction Gap is the most consequential silence in existing frameworks. A successful Dialogue names it and commits to closing it. Second, the Dialogue must establish that community-led verification is not a consultation add-on but a design requirement. AI governance is only as legitimate as the communities that participate in building it. Frameworks developed without the epistemic authority of affected communities, including Indigenous peoples, LGBTQIA2S+ communities, and Global South populations whose languages carry near-zero representation in training data, will reproduce the exclusions they claim to address. Third, the Dialogue must produce a clear signal that the governance window for AI accountability is being treated as finite; therefore, there is an urgency to finalize now. What is agreed in the next two years will determine what communities can demand for the next twenty. A successful Dialogue does not defer these decisions to future sessions but makes binding commitments that the next session can build on. A consensus should not be the signifier of success. Rather, success is whether the communities with the most at stake left with more power than they arrived with.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• AI capacity-building

These four thematic areas are not separate concerns. They are interlocking failures of the same system, and addressing any one of them in isolation will not produce durable governance. The protection and promotion of human rights is the foundation. The right to memory, the right of individuals and communities to retain ownership of their histories, identities, and cultural narratives in digital spaces, is not yet named in any international instrument as a protected right. It is already implied by existing guarantees under ICESCR Article 15, ICCPR Article 19, and UNDRIP Article 31. What is missing is the mechanism to make it enforceable against algorithmic systems that erase community memory at a scale and speed no prior governance instrument was designed to address. Transparency, accountability and human oversight are the operational requirements for that protection to mean anything. When an algorithm distorts a community's history, there is currently no obligation to disclose the training sources that produced the distortion, no pathway to challenge it, and no requirement to correct it. Oversight without transparency is performance, and accountability without correction is incomplete. The social, cultural, linguistic and technical implications of AI are where the harm is most precisely documented and least addressed. Africa is home to approximately 2,000 languages; as of 2024, leading voice AI assistants supported zero of them. Generative AI systems reproduce colonial gazes in over half of analyzed archival images. These are not edge cases; they are the structural condition of communities whose knowledge systems were never included in the data on which these systems were trained. AI capacity-building is the enabler that makes the other three actionable. Communities cannot contest distortions they cannot audit, in systems they did not help design, through processes conducted in governance spaces they cannot access.

There is one cross-cutting issue that the listed themes approach but do not name: the right to contest and correct algorithmic distortions already in circulation. Current governance frameworks address future data use. They establish consent requirements, transparency obligations, and human oversight mechanisms for AI systems being built or deployed. What they do not address is what happens after a distortion has already traveled, after a misrepresentation of a community's history has been cited in policy documents, used as the basis for institutional decisions, or embedded so deeply in a model's weights that filtering the output cannot reach the layer where the error lives. This is what our epistemic-agency researchers call the Correction Gap. It is the missing link between process rights and justice. The NOYB complaint against OpenAI filed with the Austrian Data Protection Authority in 2024 made this structural failure visible at the individual level: the system could not correct a false claim because the error was mathematical, not textual. It was embedded in the model's probabilistic weights, not in a retrievable record. If current infrastructure is incompatible with the right to truth for an individual, it is catastrophically incompatible with the right to truth for communities whose entire histories have been distorted through systems trained without their presence. Two emerging issues compound this. The first is model autophagy-the recursive loop in which AI systems trained on outputs of earlier AI models amplify existing distortions exponentially across generations, making correction progressively harder with each iteration. The second is the accelerating retirement of human content moderators, removing the last layer of contextual judgment from systems that cannot read the cultural idiom of the communities they affect most. Neither issue is captured by the current thematic clusters; yet both require urgent, named governance responses.

Innova Quantum Leap operates at the intersection of digital rights, intergenerational trauma, and AI governance, with primary focus on communities in the Global South and marginalized populations whose histories are structurally excluded from the digital record. The governance gaps we have identified are not theoretical; they are live and compounding. In Nigeria, algorithmic misrepresentation of Tiv history in Benue State has been used to deny communities their indigeneity, contributing to land dispossession, conflict, and loss of life. The digital record does not wait for affected communities to curate it. By the time those communities arrive, the narrative has already been set, already been cited, and already shaped the institutional response to their situation. Across the African continent, home to approximately one third of the world's languages, leading AI systems were trained on data in which those languages constitute a vanishing fraction of available text. The result is not merely underrepresentation, it is structural exclusion from the systems now determining what histories survive, what knowledge registers as legitimate, and what communities are rendered visible or invisible in the digital future. For LGBTQIA2S+ communities across the region, automated content moderation simultaneously over-flags legitimate expression as sensitive while failing to remove hate speech directed at those same communities. Every wrongful removal is a document of lived experience erased from the digital record of the next generation. The opportunity in this moment is also real. The African Union Continental AI Strategy and Zambia's National AI Strategy both position technological self-determination as a governance priority. The communities most harmed by current systems are also producing the most precise diagnoses of what those systems are doing and why. That precision is the raw material for governance that actually works. What is missing is the instrument that translates community knowledge into enforceable international standards.

The AI Dialogue arrives at a moment of legal crystallization. The infrastructure of AI governance—rules determining who bears liability for AI outputs, who has the right to challenge them, and which communities have standing in the processes decided—is in finalization now across three competing frameworks: the market-led model prioritizing innovation and minimal regulation, the rights-led model offering individual privacy protections within Western legal structures, and the equity-led model advocating for digital sovereignty and cultural preservation across the Global South. The communities with the most at stake in how these frameworks resolve have the fewest seats at the tables where they are being negotiated. The AI Dialogue can play a role no existing mechanism currently fills: to be the space where the equity-led model acquires the institutional weight to compete with the other two. Not by producing another declaration, but by doing three specific things. First, it can establish the Correction Gap as a named governance priority, formally recognizing that the absence of a legal pathway for communities to contest and correct algorithmic distortions already in circulation is a human rights failure requiring a binding response, instead of a future agenda item. Second, it can mandate that community-led verification is a condition of legitimate AI governance, not an optional consultation layer. This means affected communities must have binding, not advisory, authority in the governance processes that shape the systems affecting their lives. Third, it can create the connective tissue between existing mechanisms—UNDRIP, the Global Digital Compact, the EU AI Act, the CARE Principles—that currently operate in parallel without producing enforceable, cross-border accountability for algorithmic harm. The Dialogue's unique value is universality; every country has a seat. The question is whether the communities within those countries have a voice that the seat is required to carry.

Several existing initiatives have produced the evidentiary and methodological foundation that the AI Dialogue should treat as its starting point rather than rediscovering. The CARE Principles for Indigenous Data Governance, developed by the Global Indigenous Data Alliance, establish the framework for collective benefit, authority to control, responsibility, and ethics in data use. The AI Dialogue should shift the current global reference to the framework as a voluntary ethics standard by enshrining this developed expression of community data sovereignty as a legal requirement for AI training. The Kaitiakitanga License, developed by Te Hiku Media in Aotearoa New Zealand, provides the proof of concept for sovereign source data governance. It demonstrates that communities can retain jurisdictional authority over their cultural data while still enabling AI development. The AI Dialogue should recognize this model is not an exception. It is a replicable governance architecture. The Data Provenance Initiative's audit of major AI training datasets, including C4, RefinedWeb, and Dolma, documented a 25 percent decrease in available high-quality data between 2023 and 2024, alongside systematic stripping of metadata that renders community voices invisible while their content remains embedded in model weights. The AI Dialogue should mandate transparency standards that the DPI's methodology makes technically feasible. The NOYB complaint against OpenAI regarding the Right to Rectification under GDPR Article 16 established that current AI infrastructure is structurally incompatible with the right to truth. The AI Dialogue should treat this not as a European legal matter but as the opening argument for a universal Correction Gap remedy. The added value the AI Dialogue brings to all of these is jurisdiction and universality. Each existing initiative operates within a specific legal, regional, or sectoral boundary. The Dialogue is the only space with the mandate to make their combined logic binding across all of them.

The AI Dialogue will only produce durable governance if its format reflects the principle its mandate claims: that every stakeholder has a meaningful seat, not merely a visible one. Meaningful participation requires structural design, over good intentions. For governments, the contribution is jurisdictional commitment. Member States should arrive at the Dialogue with specific, named governance gaps they are prepared to close, not general statements of support for responsible AI. The Dialogue's format should require governments to respond to community-submitted evidence of algorithmic harm within their jurisdictions, creating accountability between the seat at the table and the communities that seat is supposed to represent. For civil society and community organizations, the contribution is epistemic authority. These are the stakeholders who possess the contextual knowledge to verify what algorithmic systems are actually doing to specific communities. The Dialogue's format should reserve dedicated input rounds for civil society that are not sandwiched between technical and governmental sessions, but treated as primary evidence. Testimony from affected communities should carry the same evidential weight as technical audits. For the private sector and technical community, the contribution is transparency. Developers and platform operators should be required to submit training data composition disclosures and moderation decision criteria as a condition of participation, not as a voluntary gesture. Participation without disclosure is lobbying. For academia, the contribution is independent verification. Research institutions should be tasked with auditing the gap between what governance frameworks claim to protect and what affected communities report experiencing. That gap is currently vast and poorly documented. On format, the Dialogue must invest in genuine hybrid infrastructure that centers Global South participants as primary contributors, not accommodated guests. Scheduling, language access, connectivity support, and psychosocial resources for participants sharing testimony of harm must all be designed in advance, not improvised on the day.

Underrepresentation in global AI governance discussions isn't accidental, but structural, and it maps precisely onto the communities most harmed by the systems being governed. Four groups require urgent, named inclusion mechanisms that aren't invitations, but designed pathways with resources attached. Indigenous peoples and oral knowledge communities are the most critically absent. Their knowledge systems are primarily oral, their histories were never digitized or were digitized badly, and their languages are grossly misrepresented in the training data of the systems now narrating their cultures back to them. Inclusion requires the infrastructure to participate: translation, connectivity support, and governance processes that recognize oral testimony as equivalent in standing to written submission. As a template, the Māori Data Sovereignty Network and Te Hiku Media demonstrate what Indigenous-led AI governance looks like in practice. It would be erroneous to treat them as case studies. LGBTQIA2S+ communities in jurisdictions where their existence is criminalized face a specific double exclusion: they are simultaneously over-monitored by automated moderation systems and unable to participate openly in governance processes without risk. Anonymous and pseudonymous contribution pathways, with genuine digital security infrastructure, are a precondition for their meaningful inclusion. Global South civil society organizations, particularly those working in low-resource language contexts, are routinely excluded by the cost and scheduling structures of international governance convenings. Dedicated funding streams, equitable scheduling across time zones, and formal recognition of regional civil society inputs as primary evidence rather than supplementary commentary are the minimum requirements. Finally, youth-led organizations from the Global South carry the generational argument that the Dialogue most needs to hear: the digital record being built now is the inheritance their communities will live with. Their inclusion is not symbolic; it is the only way to ensure that what the Dialogue produces is accountable to the future it claims to be building.

The formats that have historically dominated international governance convenings are panel presentations, prepared statements, and moderated Q&A. They are optimized for institutional voice, not community testimony. If the AI Dialogue is to produce governance that reflects the communities most affected by algorithmic systems, formats designed for a different kind of knowledge are crucial. Four engagement formats would make a material difference. The first is structured testimony rounds, modeled on transitional justice practice. Affected communities present documented experiences of algorithmic harm directly to decision-makers, who are required to respond within the same session. This is not a consultation panel, but an evidentiary exchange with named accountability for follow-through. Colombia's Special Jurisdiction for Peace treats collective memory as a fundamental component of justice. The AI Dialogue should borrow that logic. The second is thematic collaboration spaces organized around the Dialogue's core governance gaps rather than its thematic clusters. Each space brings together technologists, community practitioners, policymakers, and heritage holders to work on a specific, named problem: the Correction Gap, synthetic heritage, training data composition, moderation accountability, then produce a draft governance clause by the session's end. The output is not considered a recommendation. It is raw material for a binding instrument. The third is asynchronous contribution infrastructure that remains open before, during, and after the Dialogue. Communities in low-connectivity environments, different time zones, or high-risk political contexts cannot participate in live sessions. A structured digital submission process, with genuine language access, anonymity options, and a named person responsible for ensuring submissions reach decision-makers, expands the Dialogue's epistemic base without expanding its schedule. The fourth is collective witnessing—a closing format in which contributions from underrepresented communities are read into the formal record by Dialogue participants, creating shared accountability for what was heard and what was committed to in response.

The most effective AI governance examples share a common architecture: they place the communities most affected by algorithmic systems at the center of design. Three examples demonstrate what this looks like in practice, and one emerging instrument is designed to extend their logic into binding international standards. The Te Hiku Media Kaitiakitanga License, in Aotearoa New Zealand, shifts the legal status of Indigenous data from intellectual property to collective heritage. It requires users of Māori language data to demonstrate alignment with the CARE Principles and provide tangible benefit back to the iwi. Non-permissive by default and functioning as a legal barrier to corporate data extraction, it has enabled the Māori people to build accurate speech-to-text tools while retaining jurisdictional authority over their linguistic heritage. This is the most fully realized model of sovereign source data governance currently in operation. The Global Indigenous Data Alliance's CARE Principles for Indigenous Data Governance, provide the theoretical and procedural framework that the Kaitiakitanga License implements. They establish collective benefit, authority to control, responsibility, and ethics as the non-negotiable conditions of legitimate data use. Applied as a legal requirement rather than a voluntary standard, they would transform the terms on which AI systems are trained on community data globally. The Data Provenance Initiative's systematic audit of major AI training datasets provides the technical methodology for mandatory transparency. Their documentation of metadata stripping, consent violations, and linguistic imbalance across datasets demonstrates that provenance audits are technically feasible and institutionally necessary. The Ethical Digital Memory Charter, currently in community-led development by Innova Quantum Leap, proposes the governance instrument that connects these examples into an enforceable international framework-recognizing memory as a human right, mandating provenance audits, establishing community data sovereignty, and creating the legal pathway to contest and correct algorithmic distortions already in circulation.