Part 3: Advocating for AI Accountability and Governance

Around the world, civil society organisations, coalitions, and activists have been advocating for stronger AI governance, in line with human rights standards, and for mechanisms that hold developers and deployers of harmful and invasive AI technologies accountable.¹²² For example, a coalition of 150 civil society groups in Europe worked together to safeguard human rights through the development and application of the EU AI Act.¹²³ Following the adoption of this Act, civil society had to further monitor its implementation, as the European Commission and EU Parliament took steps to weaken its safeguards.^{124 125} In Brazil, civil society advocacy has been successful in addressing a weak and self-regulatory AI governance approach lobbied for by the private sector.¹²⁶ In the Occupied Palestinian Territory, collaborative media investigations revealed the extent of Israel’s lethal deployment of AI and biometric surveillance systems in connection to the genocide in Gaza¹²⁷ and the complicity of specific tech companiesin these abuses and war crimes.¹²⁸ In India, the Internet Freedom Foundation has been mapping FRT projects in the country¹²⁹ and pushing for transparency in the deployment of AI in public spaces.¹³⁰

Most survey respondents (nearly half) stated that their organisations do not conduct any advocacy work to influence how AI systems are developed, deployed or regulated. About a third of respondents stated that they would like to influence how AI systems are developed, deployed or regulated. Only 17 percent of respondents stated that their organisations seek to influence the ways AI systems are developed, deployed, and regulated.

There are obstacles that stand in the way of broader civil society participation, particularly in the Global South, and contribute to the debate around AI governance and accountability. The main obstacles are a lack of resources, a hindering political and regulatory environment, and a lack of transparency surrounding the development and deployment of AI systems.

First, in terms of resources, there is the lack of funding and staff with needed knowledge and skills in their organisations. These two are interrelated as the lack of funds affects organisations’ ability to hire employees and consultants who can support them with AI accountability and governance activities.

A Ugandan organisation, R0048, said that it “lacks the necessary expertise in AI governance, policy and technical development, making it challenging to effectively advocate for responsible AI practices”. It further explained in the survey response: “Limited financial resources hinder the organization’s ability to participate in key forums, workshops and conferences on AI governance and regulation, restricting their ability to engage with policymakers, experts and other stakeholders. The organization requires training and capacity-building programs to enhance its understanding of AI technologies, their applications, and implications for human rights and social justice”.

“…Internally, we face a lack of technical knowledge and skills needed to fully understand how AI systems function, how they affect human rights and how to design strong advocacy strategies around them. This gap makes it difficult to analyze or respond to the growing use of AI in surveillance, information control and public communication”, said a respondent from Tanzania (R0201).

In other cases, some respondents stated that there is a general shortage of people with the needed knowledge and technical skills in the contexts in which they are working. “The main obstacle is severe talent scarcity in Nigeria. We lack local experts to technically audit proprietary AI systems for bias and non-compliance. This knowledge gap means we cannot credibly challenge powerful deployers, undermining our advocacy for fairness and accountability in critical areas like facial recognition”, one respondent who chose to remain anonymous said.

In some cases, available funding tends to exclude those conducting critical AI work, particularly when that funding is coming from the tech industry. Disha Verma, Program Manager at the Global Tech Institute asserted:

“I will say that there is a glaring lack of funds right now to fund critical work on AI because there is way too much money to do favourable work on AI. If you want to work closely with an AI provider, and build an AI safety institute, or have red lines and guardrails for AI, work with the government on a regulation, the positive reinforcements are many. You will get a lot of money from it, from all kinds of funders coming in from outside India, or within India. But the minute you start to say that maybe we do not need AI right now, maybe we do not need AI in this sector, or maybe we do not need AI in policing, we do not need AI in welfare services, people cut you out. There is no money for that research. So how do we continue making that knowledge? How do we continue creating knowledge and research and outputs and advocacy material when we are not getting paid as researchers and activists? So that is really concerning to me. Also to note that AI has become a tool of soft power for India. We take a lot of nationalistic pride in our AI systems and our AI mission. So, the minute you start to question it, it becomes like an anti-national sentiment and a statement that can be jarring for the government, so they try to shut you off and basically attack you”.

Another major hurdle for civil society participation in shaping how AI is developed, deployed, and regulated is an unfavourable regulatory and political environment. Respondents and interviewees noted, in particular, the lack of robust regulations on AI that align with international human rights standards in addition to a climate that hinders civic participation, further complicating advocacy efforts. Despite repaid and disruptive advancements in the AI field and increased adoption across different sectors and industries, robust human-rights based regulation of AI is still moving at a slow pace. In 2024, the EU adopted the AI Act, a comprehensive framework for regulating AI that takes a risk-based approach.¹³¹ The Act bans applications with unacceptable risk and imposes strict requirements for high-risk applications such as FRTs and emotion recognition systems. It only imposes transparency requirements for lower-risk systems. The Act, however, has major loopholes, including broad exceptions for use of high-risk AI systems for the purposes of defence, law enforcement, and migration.¹³² Furthermore, under pressure from the tech industry and the administration of Donald Trump in the U.S., the EU Commission proposed to delay the implementation of some provisions in the AI Act and weaken some of its safeguards.

In other regions, Brazil has been in the process of drafting a similar law that also adopts a risk-based approach but it has yet to be adopted. Other countries, like Japan¹³³ and South Korea,¹³⁴ have adopted a ‘soft approach’ in their AI legislation, prioritising innovation and imposing limited obligations on deployers and developers compared to the EU AI Act. Most recently, India adopted issue-specific rules on deepfakes which mandated the labelling of deepfakes and banned “deceptive impersonations, non-consensual intimate imagery and synthetic content tied to serious crimes”.¹³⁵ It also mandated platforms to take down infringing content within three hours.¹³⁶ This short deadline was criticised by digital rights groups which argue this will only result in more automated removal decisions and could have a chilling effect on freedom of expression.¹³⁷

Many countries are currently opting to adopt non-binding principles or codes such as India,¹³⁸ Saudi Arabia,¹³⁹ and Singapore.¹⁴⁰ There are also guidelines adopted at the international or regional level such as the OECD’s Recommendation of the Council on Artificial Intelligence (the “Recommendation”)¹⁴¹ and the Continental African Intelligence Strategy.¹⁴²

There are “weak and sometimes non-existent regulatory frameworks in Kenya and much of Africa; there is a lack of comprehensive AI governance laws as well as enforcement mechanisms that protect communities from misuse of AI. Existing data protection laws (like Kenya’s Data Protection Act) may not be fully enforced as well as adapted to the complexities of AI. There are often regulatory capture and lack of independence in oversight bodies”, said one respondent, adding that as a result, “CSOs have limited legal and institutional pathways to challenge unethical and harmful AI deployments”.

Lack of political willingness and an unstable political environment were also cited as factors that hinder adoption.^{143 144}Civil society in many countries is working in restrictive and repressive operating environments. As a result, they cannot effectively question powerful actors’ deployment of AI to hold them to account. Freedom of expression and access to information, freedom of peaceful assembly and the right to privacy are often tightly controlled. According to the CIVICUS Monitor’s 2025 data, “Only 7.2 percent of the global population now lives in countries where civic space is open or narrowed, 7.5 percentage points less than in 2024, indicating a further deterioration of civic space conditions globally”.¹⁴⁵

There is a “fear of retaliation or repression raising concerns about government-deployed AI systems, especially those related to surveillance, censorship, and policing, which can be politically sensitive. Civil society organisations may fear retaliation, de-registration or intimidation when calling out unethical practices. This inhibits open advocacy and transparency, particularly in restrictive civic spaces”, one anonymous respondent stated. Another respondent working in Afghanistan wrote: “Our organisation does not currently engage in direct AI advocacy due to operating restrictions and security concerns in Afghanistan. The lack of resources, expertise, and safe civic space limits our ability to influence AI policies or participate in related forums. However, we are aware of the human rights implications of AI and hope to engage more once a safer environment becomes available”.

Additionally, algorithmic opacity makes it difficult for civil society to understand how AI systems are deployed, hindering their ability to hold deployers and developers to account. In fact, 45 percent of respondents listed the lack of transparency from deployers on when and how they deploy these systems and 38 percent listed the lack of algorithmic transparency from developers on how their systems work as obstacles to exercising oversight and calling for accountability.

According to R0201,“Externally, there is a lack of transparency from government and private sector actors regarding when, where, and how AI systems are deployed. This secrecy prevents civil society from monitoring potential abuses, assessing compliance with human rights standards, or holding deployers accountable. Together, these challenges limit our ability to meaningfully influence AI governance and advocate for responsible and ethical use of AI technologies”. Another respondent stated that “many AI systems, especially those developed by private companies or deployed by governments, operate as ‘black boxes’ with little to no public visibility into: 1. How they are built 2. What data they use 3. What assumptions or biases they encode. It becomes nearly impossible to assess and even challenge harmful impacts, especially when those affected lack the technical knowledge as well as legal means to demand redress”.

In other cases, governments’ own communication about when AI is deployed and when it is not can be confusing to advocates, as Verma noted of the Indian government’s approach in an interview. “The government is not honest about what is AI and what is not AI. They call everything AI. We do not know if it is just machine automation, if it is just simple maths. But when you ask, when you seek further questions, it turns out to be just a thermal imaging system, for instance, not AI. It is called AI for the hype train to add to the nationalistic fervour and so on and so forth”.

According to Juan Manuel García, an Argentinian consultant at the intersection of technology and human rights, some organisations in Latin America had previously submitted Freedom of Information Requests (FOIAs) to government agencies to understand how they deploy AI, but those requests have not always resulted in positive responses. “One of the main problems that have been very common for many organisations in Latin America is access to information regarding the use of AI. The Argentinian government in mid-2024 created this new agency that is called the Unit for AI for Security that has a really broad spectrum of activities and some of the activities are based on the use of AI surveillance of social media. Different organisations have presented FOIA requests, but they did not answer them, so we do not know how they are using it…Sadly, we do not know the specificities, because they do not answer the requests. It is a problem”.

Lacking resources, access to information, and facing retaliation for their work, civil society is disadvantaged compared to powerful and well-resourced government and private sector actors in shaping how AI is deployed and governed. Civil society also faces exclusion from national and international AI discussions.

“Civil society organisations like ours often face closed-door policymaking and technical barriers in understanding AI systems. Without funding and open access to information, it is hard to hold large tech actors accountable or influence regulation effectively”, said a Kenya-based organisation (R0168).

“There are few formal mechanisms for civil society participation in policymaking, and many discussions on AI governance take place behind closed doors, limiting NGO influence and representation”, said disability rights organisation Spring NGO (R055).

This often results in a power imbalance, where the private sector with a few large players and governments have more power to influence these discussions and spaces and their outcomes. The lack of representation of civil society means human rights and impacts on excluded groups are neglected.

“We are quite involved in AI governance; we talk to our industries whenever we can, in terms of how you govern AI, how do you set up guardrails, etc. AI safety is a big thing in India. But the problem is that it is really noisy, the space. There are so many conversations happening. There is a bustling and booming AI industry in India, which wants in on every room, which wants to say that innovation versus regulation, we need innovation, we do not need regulation. And then there are the government quarters and the tech conservatives that say that, no, regulate everything, put everything to the ground…The problem is that our government is authoritarian, so giving them powers through legislation and giving them powers through regulation is dangerous because they will surveil, they will intercept, they will do all of that and curb free speech and right to privacy whenever they can. But on the other hand, letting the private sector run amok and do whatever they like is also dangerous, because these then become very powerful, and as consumers, it tips the scale of balance of power quite a bit. And we are not equals, we are not able to hold our platforms accountable. So either way, it is a lose-lose situation for the citizens”, said Verma of the Tech Global Institute.

While civil society in Brazil has been successful in pushing back against an initial AI draft bill that “was really favored by the industry”¹⁴⁶ and had “a very light touch and minimalistic approach”¹⁴⁷ to regulation, their resources still do not match those of the tech industry. As a more comprehensive draft bill currently in consideration, Bill No. 2338/2023, has yet to be finalised and enacted, the private sector has increased pressure on lawmakers.¹⁴⁸

Civil society from the Global South faces additional hurdles participating in international discussions and forums on AI governance such as the global AI summits¹⁴⁹ and the UN Global Dialogue on AI Governance¹⁵⁰ due to expensive travel costs and strict visa rules.

“Civil society voices from the Global South are rarely present in major AI policy spaces. Even when invited, discussions are often dominated by private sector actors. Without resources to participate, small organizations like ours are left to adapt to systems we had no say in shaping”, said an organisation working in Nigeria (R0051). Another pointed out the power imbalance that advantages companies and governments: “Large tech companies and government institutions often have asymmetric access to resources, expertise and policy influence. Civil society organisations, especially those working at grassroots level, may lack the technical capacity to audit systems and the political capital to influence AI policy. This creates a knowledge and power gap that prevents meaningful accountability and engagement”.

The dominance of the English language in these spaces is another hurdle for voices from the Global South. “Civil society voices from the Arab region are rarely included in global AI policy debates. We aim to represent these voices but face linguistic and logistical limitations”, said an organisation based in Jordan (R0114).Furthermore, there is the lack of digital remote participation for those facing visa restrictions or cannot afford travel. In some cases, where remote participation is available, lack of connectivity or expensive data bundles hinder online participation. “Civil society in South Sudan is often excluded from AI policy discussions due to limited connectivity and resources. We need access to regional and international policy dialogues to ensure local realities are considered in global AI ethics debates”, said one respondent, R0035.

According to R0171, “most AI discussions are led by actors in the Global North. Local CSOs like
EAO rarely have access to consultations where African civic perspectives can influence AI ethics or governance frameworks”.