Guarding Democracy in the Age of AI: When Everyone Has to Do Their Part

wall of surveillance cameras

By Vija Kalniņa, PhD, Business AI Ethics Initiative Lead Researcher and Lily Noyes, Business AI Ethics Initiative Research Assistant

Introduction

As the development of artificial intelligence (AI) technologies continues to create seemingly endless opportunities, it also introduces significant dangers as these models and algorithms become deeply intertwined with the functioning of both social and political life. According to Sanford HAI 2025 AI Index Report, governments are increasingly developing and establishing AI policies and regulations. For them this process comes with many wide-ranging challenges – from regulating a rapidly developing technology with many unknown aspects and uncertainties to finding a middle ground between various pressures arising from competing interests.

One of the main challenges for governments is threats that AI poses to state orders, whether they are democracies or autocracies, with democracies encountering more serious risks than any other state governance system.

This blogpost explores the main threats that AI poses to democracies and fundamental rights, how governments in different democracies develop their policies, and to what extent concerns about AI’s negative impact on democracies are taken into consideration. Finally, we explore whether other non-governmental players could become guardians of democracy and fill the gaps that governments have created. We believe that the challenges AI poses to democracies are complex enough that there is a role for every individual and legal entity to play.

1. Threats that AI Poses to Democracy and Fundamental Rights

Over the past several decades, scholars, thought leaders, and various international institutions and NGOs have increasingly voiced concerns about the negative impact of algorithms and AI on democratic processes such as freedom of expression or the integrity of fair and open elections (see e.g. Freedom on the Net 2018 by Freedom House).

As summarized by Allen and Weyl in their comprehensive article “The Real Dangers of Generative AI,” an essential concern is the potential for AI systems to interfere with democratic processes through widespread deception, surveillance, and manipulation of information. These developments undermine key democratic principles of authentication, transparency, and privacy. Furthermore, the growing influence of AI technologies risks concentrating power in the hands of a limited group of actors such as technology experts or financial institutions who operate with minimal oversight or transparency, which shifts away from a democratic model.  Allen and Weyl argue that the strength of democracy depends on human collective intelligence and open participation, which are both threatened when decision-making becomes centralized in opaque technological systems. They frame the danger of AI as a tension between social fragmentation and domination by a “technological singularity” that risks destroying democratic governance systems. 

Beyond the threat to democratic institutions, AI systems also endanger fundamental human rights protected under various international and regional normative frameworks (such as the UN, OECD, or Council of Europe). As recognized by many organizations, such as UNESCO, algorithmic bias and discrimination threaten the pluralism and diversity essential to these rights, highlighting the potential of AI systems to erode the fundamental human rights of a wide range of populations.

2. Government Approaches to Safeguarding Democratic Values

The responsibility to protect democracy and fundamental rights naturally falls on governments. Overall, it appears that governments of democratic states are aware of the potential risks posed by AI. However, when crafting policies and adopting regulations, governments must balance the need to safeguard society with other interests – most notably, the drive to foster innovation and reap economic benefits. Approaches differ across jurisdictions: some governments favor minimal regulation and maximum freedom for innovators, while others take a more cautious stance, clearly establishing what risks are unacceptable and drawing red lines that should not be crossed in AI development or deployment.

The following examples of AI policies in different democracies illustrate what direction each jurisdiction leans towards in their attempts to find balance between progress and caution, and further explore what challenges come with each chosen approach:

  • When it comes to AI, the UK has always positioned itself as pro-innovation while also emphasizing its strong commitment to protecting democratic values through embedding them in international AI standards and regulatory intervention to address threats to democracy. However, in 2025 the AI Opportunities Action Plan shifted the focus to innovation and growth and softened its statements on AI regulation, aiming to achieve AI safety via the AI Safety Institute and a proportionate, flexible regulatory approach.

  • The regulatory approach of the United States at the federal level has varied largely over the last five years as aligned with changing presidential administrations. Published in July 2025, America’s AI Action Plan frames AI as an economic and national security imperative that largely promotes innovation over caution. While the Plan addresses concerns relating to privacy, security, misinformation, and model performance and reliability, it leaves gaps in areas such as fairness and human oversight that are intertwined with the safeguarding of fundamental rights. Furthermore, some voice concerns that supposed protections of free speech within the Plan are closely tied to political agenda and ideology.

  • On the state level, legislation does not always follow the innovation-centric position of the federal government. For example, California passed the Transparency in Frontier Artificial Intelligence Act (SB 53) in September 2025, making it the first U.S. state to pass a law that imposes transparency, safety, and accountability requirements on certain “frontier” AI developers to enhance public safety and oversight of advanced AI systems.

  • Canada is adopting a gradual, precautionary approach to AI governance. Within the government itself, Canada places responsibility, ethics, and commitment to law at the forefront of its adoption and use of AI within the government sector through tools such as the Guide on the Use of Generative Artificial Intelligence, Algorithmic Impact Assessment Tool, and the Directive on Automated Decision-Making. However, the proposed Artificial Intelligence and Data Act, which sought to ensure the deployment of safe and non-discriminatory AI systems as aligned with protecting human rights interests, failed to be passed and emphasizes the slow-moving progress of Canada’s government to impose national AI governance legislation. 

  • Japan has recently adopted an industry-led, innovation-driven approach, as evidenced by the passing of the AI Promotion Act in May 2025. This represents a shift from previous frameworks such as the AI Guidelines for Businesses Ver 1.0 policy document and the Interim Report from Japan’s AI Policy Study Group which emphasize regulation and adherence to standards of human rights, democracy, safety, fairness, and the rule of law when developing and implementing various AI systems. 

  • While Australia currently has no direct laws governing AI, the government advocates for an AI landscape that aligns with their AI Ethics Principles. These eight pillars advocate for increased safety and reliability of AI systems and seek to reduce AI-related risks to humans and government institutions. However, the principles are based on voluntary participation rather than legal requirements and leave room for further future legislation. Within the government itself, the Technical Standard for Government’s Use of Artificial Intelligence maintains AI safety standards aligned with the ethics principles while allowing AI innovation within the governmental sphere.

  • The EU has approached AI policy by placing the protection of fundamental rights enshrined in the Charter, including democracy and the rule of law, at its center of it. These goals, as stated in Article 1(1) of the EU AI Act, are the core foundation of this regulation. However, the EU has received harsh criticism for creating a very complex, hard-to-interpret and enact, fragmented, and premature legal framework from industry and in the Draghi report.

Apart from these government-level initiatives, the Council of Europe Framework Convention on AI should also be noted as a serious attempt to bring together democratic governments and agree on ensuring AI development and deployment that is aligned with human rights, democracy and rule of law. While the number of signatories is promising, the national discretion in interpretating the Convention, disclaimers and enforcement mechanisms shed a shadow of uncertainty over it.

Depending on the context, the same government approaches face both criticism and praise (e.g. the EU is praised for truly protecting fundamental values but is criticized for stifling innovation). However, the true effectiveness of these strategies in achieving their intended goals remains to be seen, as it will take time to observe their full impacts and outcomes. In the meantime, these examples highlight the challenges that governments encounter as they strive to strike the right balance among competing interests.

3. The Role of Nongovernmental Players in Safeguarding Democratic Values

There might be several reasons why governments could fail to govern AI in a way that secures not only economic benefits but also the protection of democratic values and fundamental freedoms. Issues could arise not only from failing to strike the right balance between different interests, but also from the creation of legal frameworks with inherent flaws that prevent effective enforcement.

Regardless of the cause, this raises an essential question of who – individuals, NGOs, academia or businesses – is in the position to step in and address the gaps left by governments?

  • Individuals: Individuals have capacity to influence how AI is deployed by choosing not to use it for unlawful or unethical purposes, and by avoiding uses that harm democracy or humans. They also play a key role in raising concerns and putting pressure on governments and businesses – whether as employees (including whistleblowers) or as consumers – to create and implement safe, trustworthy AI.

  • NGOs: NGOs have the capacity to mobilize and represent the interests of larger groups. They can exert pressure on businesses and governments, encouraging them to take responsible action regarding AI development and use.

  • Academic & Research Institutions: Researchers and scholars can highlight potential threats to democracy by supporting their concerns with evidence-based research. They can also propose research-driven solutions for both technological and policy challenges related to AI.

  • Businesses that develop AI: These companies are uniquely positioned to design AI systems that align with democratic values and to build in safeguards that by design limit or prevent harmful uses of AI.

  • Businesses that deploy AI: These companies can consciously choose to deploy AI in a trustworthy and safe manner. They can also encourage developers to produce AI systems that support these attempts. By establishing best practices in the industry, these companies can additionally pressure both competitors to follow these standards and governments to adopt policies that approve and support these approaches.

Each of these players occupies an important role toward ensuring AI is developed and deployed in ways that do not harm democracy or society. However, each stakeholder faces limitations regarding what they can achieve and the resulting scale of their impact. Factors such as financial resources, the breadth of influence, and commitment to doing the right thing all affect how much – and for how long – each group could contribute. Moreover, none of these stakeholders besides governments themselves possess the legal authority to act against malicious actors who intentionally develop and deploy AI for harmful purposes.

Relying solely on these stakeholders would certainly yield some positive outcomes, but there is concern that certain challenges may be beyond their capacity to address. Additionally, this approach is not sustainable in the long term as businesses struggle with the balancing between competitiveness and responsible caution.

Conclusions

Even before AI became a mainstream topic, the Presidio Institute suggested cross-sector collaboration to be the most fitting approach for solving large-scale challenges in an increasingly complex and interconnected world. 

Understanding the challenges that governments face in protecting democracy and witnessing their struggles with formulating AI policies and establishing effective safeguards, cross-sector collaboration between governments and non-governmental players might be the only approach that will allow sustainable long-term solutions. The thesis that this could be a fitting approach to AI challenges is supported by the fact that governments are already increasingly exploring these options not only for AI research or infrastructure development, but also by co-drafting codes of best practices or involving stakeholders in consultations about the regulatory frameworks.

We believe that such collaborations should be expanded and could combine the distinct strengths, methods, and perspectives of various stakeholders, creating an added layer of security where certain parties are able to catch issues and prevent AI incidents in areas where another fails or overlooks.

 

*** 

  • Jeffrey Saviano: Business AI Ethics Initiative Leader; Edmond & Lily Safra Center for Ethics, Harvard University | jeffreysaviano@fas.harvard.edu 

  • Jonathan Hack: Director of Content & Strategy; Edmond & Lily Safra Center for Ethics, Harvard University 

  • Vija Kalnina, PhD; Business AI Ethics Initiative Lead Researcher 

  • Lily Noyes; Business AI Ethics Initiative Research Assistant 

  • Ryan Wettre; Business AI Ethics Initiative Research Assistant 

 

The Business AI Ethics research team is part of the Edmond & Lily Safra Center’s ongoing effort to promote the application of ethics in practice. Their research assists business leaders in examining the promise and challenges of AI technologies through an ethical lens. Views expressed in these posts are those of the author(s) and do not imply endorsement by ELSCE.