The Missing Foundation in AI Governance: Building Trust Across Parties
By Rebekah Tweed and David Ryan Polgar
Artificial intelligence product releases continue to drop at a breakneck pace, along with the increasing capabilities of the most up-to-date frontier models they’re built upon.
From generative multi-modal models to emerging agentic systems that take actions on our behalf, myriad forms of AI are rapidly reshaping how we live and work. Yet while innovation surges ahead, governance frameworks lag—fragmented, inconsistent, and often reactive.
The result? A growing gap between technological capability and societal safeguards. Something needs to change. The yawning gap between the speed of innovation and the slowness of societal consideration is problematic, as it leads to a situation where technology is thrust upon the public without their implied consent regarding its usage.
Collective Understanding, Involvement, and Action
Here at All Tech Is Human, we’ve spent years building a broad community of responsible technologists and changemakers across civil society, government, industry, and academia to confront exactly these types of challenges. Our community weaves together a broad range of backgrounds and disciplines, which is advantageous for understanding values, tensions, and tradeoffs requests to tackle complex problems. We specialize in creating a conducive environment for holistically understanding an issue, with a powerful network that can be quickly mobilized. Key to reducing the major gap between innovation and consideration is having an approach that can move at the speed of tech.
We wanted to apply our network to understanding the concerns of AI Governance professionals. So throughout 2025, we gathered insights from 275 individuals through our AI Governance interest form and hosted workshops on AI Governance in New York and AI Assurance in London.
The input we received derived from individuals working at companies, universities, and organizations such as Amazon, Amnesty International, Barclays, BBC, BCG, Cohere, Columbia University, Deloitte, Department of Homeland Security, Ericsson, Fordham, Georgetown University, Google, Harvard Kennedy School, Harvard University, Humane Intelligence, IBM, Integrity institute, Intel Labs, LinkedIn, London School of Economics and Political Science, McKinsey & Company, Meta, Microsoft, Nokia, NYU Center for Global Affairs, Omidyar Network, Pace University, TikTok, Trustible, University of Bristol, University of British Columbia, University of Calgary, University of Michigan, University of Oxford, Yahoo, York University, and more.
When asked what most concerns them about AI governance, the responses were strikingly consistent: from regulatory gaps to declining public trust, ten themes emerged, painting a clear picture of what’s at stake if we fail to act.
Regulatory Lag and Fragmentation
Political Pressures
Global Governance Challenges
Concentration of Power
Accountability Gaps
Individual Harms
Societal Scale Impacts
Complexity Outpacing Governance
Public Engagement and Literacy Gaps
Implementation Gaps
1. A Fractured Landscape
The most pressing concern we discovered from our feedback is regulatory lag and fragmentation. The global tech industry's race to deploy the most capable AI models is resulting in technological advancements that develop far faster than any governance structures can adapt. In addition, different jurisdictions from the U.S. to the UK, and from the EU to China, to various nations throughout the Global South are taking divergent approaches, creating confusion and enabling industry actors to engage in “jurisdiction shopping.” Powerful players exploit loopholes, while vulnerable communities shoulder the risks.
This fragmentation undermines trust. In this environment, governance is only as strong as its weakest link, and without cohesive, timely structures, public faith in the legitimacy of AI systems will continue to erode.
“The regulatory landscape is maturing methodically, but cannot keep pace with the rapid deployment of increasingly complex AI systems. This lag creates both risk and opportunity. In this context, AI assurance serves as a proactive safeguard and practical mechanism for mitigating risk, demonstrating accountability, and enabling confident innovation.” AI Assurance Ecosystem Workshop Summary Report (London)
2. Political Pressures and 3. Global Tensions
U.S. AI policy, amid current political pressures, is increasingly shaped by national security interests and global competition. Ethics and safety risk being sidelined as an “AI arms race” mentality prioritizes speed and dominance over responsibility. Meanwhile, attempts at global coordination are hampered by starkly different philosophies: China’s agile model, the EU’s methodical legislative process, and the U.S.’s market-driven approach.
These global tensions are a concern because building inclusive global standards requires trust and collaboration; without global standards, AI governance risks being both inequitable and ineffective.
“Different countries have distinct regulatory requirements and cultural values, making it difficult to create a one-size-fits-all approach to AI testing and evaluation. This is particularly challenging when dealing with issues like fairness, which may be interpreted differently across regions.” - AI Governance Workshop Summary Report (New York City)
“I'm most concerned about the level of deliberation within the governance community when the political conditions aren't conducive to formal AI governance policy in the US.” - Jeba Sania (AI Safety Researcher, Harvard Kennedy School)
4. Concentration of Power
Another recurring concern is the growing concentration of AI power. Big Tech firms, wealthy nations, and elite institutions often dominate decision-making, while marginalized communities and the Global South are often excluded. As diversity, equity, inclusion, and accessibility (DEIA) commitments wane, governance risks become a top-down process decided by the few rather than shaped by the many.
This concentration of power doesn’t just raise equity issues; it threatens the legitimacy of governance itself.
“The concentration of power, not only in the hands of Northern Big Tech, but also in the hands of Northern non-profits, NGOs and think tanks driving discussions on AI governance and ethics, and policy.” - Michael L. Bąk (Senior Non-Resident Visiting Fellow - Cyber Policy, NYU SPS Center for Global Affairs; Global Advisor, Asia at the Ethical AI Alliance)
5. Accountability Gaps
The public sees AI systems that are often opaque, untested, and deployed without clear accountability. Few mechanisms exist to hold developers, deployers, or users responsible when harms occur. This lack of clarity is especially troubling in high-stakes contexts like finance, healthcare, and law enforcement.
Trust in technology cannot be demanded – when trustworthiness must be demonstrated, trust is subsequently earned. Trustworthiness requires transparency, structured accountability, and active public engagement. Without those elements, mistrust will deepen, further widening the gap between communities and the institutions deploying AI.
“I’m concerned that some AI labs are underestimating the long-term societal backlash that could result from failing to engage meaningfully with the public on AI’s risks in the short-term. If companies avoid open dialogue and dismiss public concerns, they risk fostering mistrust that could lead either to aggressive regulatory pushback restricting the deployment of AI products and services or increased lobbying from AI labs to force adoption. It’s crucial that labs proactively collaborate with civil society now to avoid adversarial dynamics later.” - Alex S. (Fmr. Strategic Partnerships & Business Strategy Lead, USAID)
“Many consumers, workers, and community stakeholders perceive the risks of AI as outweighing the benefits…Participants largely agreed that trust must be demonstrated and maintained through transparent, structured, and ongoing assurance practices. AI assurance offers a powerful toolset for building public trust, allowing organizations to move beyond reactive compliance and begin actively addressing the concerns of the people their systems affect.” - AI Assurance Ecosystem Workshop Summary Report (London)
“One of the central issues in AI governance is the question of accountability: who is responsible when an AI system is misused or causes harm? Current frameworks often fail to allocate clear responsibility or define liability regimes which leaves a significant gap in addressing the wrongful use of AI models.” - AI Governance Workshop Summary Report (New York City)
6. Harms to Individuals and Communities
Perhaps the most deeply resonant concern is the way AI entrenches bias and exacerbates inequities, harming individuals and communities. Without robust and continuous oversight, AI systems can disproportionately harm women, BIPOC, immigrants, and low-income communities. This is particularly problematic in hiring, policing, and social services, where decisions, automated or not, are especially impactful. Meanwhile, declining DEIA commitments and limited representation in AI design mean the very groups most affected are often the most excluded from tech’s decision-making tables.
“Another challenge arises in defining what constitutes “good enough” when it comes to model fairness. Engineering teams often ask whether a model has been “debiased,” or if a model has been checked and updated to reduce bias, but answering this question is far from straightforward. Fairness is highly context-dependent, making it impossible to establish a one- size-fits-all metric. As a result, developers face uncertainty about whether their efforts to reduce bias are sufficient for model deployment. The lack of standardized metrics for fairness complicates efforts to responsibly govern AI, leaving developers grappling with the ambiguity of how to ensure ethical outcomes.” - AI Governance Workshop Summary Report (New York City)
7. Societal-Scale Risks
AI doesn’t just affect individuals; it affects entire systems. Its energy and resource demands threaten climate goals, with the Global South bearing disproportionate burdens. At the same time, advanced agentic AI systems raise the risk of widespread job displacement. And behind the scenes, the hidden labor of annotators and content moderators continues under precarious and overlooked conditions.
These societal-scale risks across climate, labor, and economic disruption demand governance frameworks that extend beyond technical safeguards to include justice and sustainability.
8. Complexity Outpacing Governance
As AI grows more complex, so do the risks. This growing complexity is outpacing AI governance as generative, agentic, and autonomous systems are being released without adequate safeguards. Misuse in surveillance, warfare, and disinformation is already a concern. But the greatest blind spot may be multi-agentic systems, where multiple layers of technology interact in nuanced ways.
Current governance structures, designed for isolated systems, are ill-equipped for this new complexity. Without intervention, the risks could quickly outpace our capacity to respond.
“[The] AI landscape is changing rapidly, not just in scale but in kind. What once seemed complex, such as facial recognition technology or basic predictive models, now appears relatively straightforward in comparison to current and incoming waves of non-deterministic, multi-modal, and increasingly autonomous agentic systems....What does fairness look like in a voice-enabled agent making real-time decisions across languages, geographies, and contexts? What does safety mean in a system that can reason, execute actions autonomously, or interact across APIs and third-party tools?” - AI Assurance Ecosystem Workshop Summary Report (London)
“I'm concerned about the lack of safeguards with some of the products being deployed, with the move to be more ‘bold’ than responsible in the US particularly, and the lack of broader education about AI responsibility and mitigations that can be put in place. However, I do see hope through organizations such as ATIH and others, and that keeps me going!” - Kerry Barker (Head of AI Governance, Sony Interactive Entertainment)
“The space seems to be very static in many places, using traditional governance and audit perspectives, which is not dynamic enough to address some of the key risks associated with AI systems.” - Chris Jefferson (CTO/Co-Founder, Advai)
“Understanding the hype vs reality around agentic systems - there is a lot of conflicting information, and that makes it hard to figure out what is really going on.” - Anastassia Kornilova (Director of Machine Learning, Trustible)
9-10. Engagement and Implementation Gaps
Finally, our community pointed to gaps in public engagement, AI literacy, and implementation. Governance decisions are too often made without meaningful public consent. Policymakers, boards, and the public frequently lack the knowledge to provide effective oversight, creating implementation gaps. Meanwhile, smaller organizations struggle to access practical governance tools, and policies often lack “teeth,” devolving into optics or “ethics washing.”
“A fundamental challenge in deploying AI responsibly is that many organizations don’t fully understand what they’re adopting, let alone how to evaluate its performance or risk. Across sectors, AI is moving faster than internal capacity to manage it, which results in a persistent knowledge gap: organizations are investing in tools they don’t fully understand, asking incomplete or misdirected questions, and struggling to connect AI use to broader strategic goals. AI assurance provides a structured solution to this challenge, not only as a technical function, but as a catalyst for increasing organizational AI literacy...Importantly, AI assurance also creates a mechanism for board-level and executive clarity. Boards increasingly want to understand the organization's risk posture, not just whether AI is being used but whether it is aligned with risk appetite and expected ROI. Assurance enables risk tiering, linking levels of exposure to specific use cases and offering leadership a clearer picture of what risks are acceptable and which are not.” - AI Assurance Ecosystem Workshop Summary Report (London)
“Organizations are still struggling to understand who owns or oversees AI governance, which can hinder responsible adoption.” - John Heflin Hopkins-Gillispie (Director of Policy and Product Counsel, Trustible)
“I am concerned that Safety can be easily considered as a blocker (and hence treated as a checkmark) by business and product teams, given the competitive business environment in AI space. Conversations on safety should involve core business decision makers to truly embed safety into the product from its inception to deployment, not be siloed amongst professionals who are already aware and working on safety. I would like to discuss how we can find creative ways to incentivize business and product teams to account for Safety from the entire lifecycle of AI models and GenAI applications, given all the known challenges and significance of safety.” - AI Policy Manager at a global tech company
The Missing Foundation: Trust
Across all these themes, one thread ties them together: trust.
Trust is the missing foundation in AI governance. Without it, even the strongest policies risk collapse. Without it, marginalized communities remain excluded, public skepticism grows, and legitimacy crumbles. Trust cannot be outsourced or assumed. It must be intentionally built, maintained, and demonstrated across the entire AI ecosystem.
That’s why All Tech Is Human is launching the next phase of our work: building a framework for an inter-party system of trust in AI systems. This framework will help bridge divides between developers, deployers, regulators, and the public. It will create structured ways to demonstrate accountability, ensure transparency, and include diverse voices in shaping the future of AI.
The future of AI governance will not be defined by any single company, government, or community. It will be defined by our ability to work together to build systems that people trust.
We invite you to join All Tech Is Human as we co-create this framework for trust. Together, we can ensure that AI systems are deployed responsibly, inclusively, and transparently, anchored in the values of accountability, fairness, and human well-being.
