B Explanatory memorandum
by Ms Deborah Bergamini, rapporteurNote
1 Introduction
1. Digital technologies have developed
more rapidly than any other innovation in human history.
Note The pace of progress has become
so fast that national and international attempts to regulate their
use, harness their benefits and mitigate the threats they pose to
society are in constant danger of becoming prematurely obsolete.
2. This is particularly true in the case of Artificial Intelligence
(AI). The use of AI systems raises concerns, in particular because
of their potentially negative impact on fundamental freedoms and
democratic processes and institutions.
3. Implications include the violation of citizens’ privacy; the
use of their personal data to control, coerce, censor or punish
their decisions and behaviour; the spread of misinformation and
disinformation to influence their opinions and choices; and the
risk of cyber-attacks on sensitive public websites and databases,
among other issues.
4. There are also significant implications for what concerns
the democratic security and sovereignty of countries, as AI tools
can become effective weapons in the hands of malicious foreign parties
who seek to destabilise democracies.
5. As stated in the International Scientific Report on the Safety
of Advanced AI, however, “AI does not happen to us: choices made
by people determine its future. The future of general-purpose AI
technology is uncertain, with a wide range of trajectories appearing
to be possible even in the near future, including both very positive
and very negative outcomes. This uncertainty can evoke fatalism
and make AI appear as something that happens to us. But it will
be the decisions of societies and governments on how to navigate
this uncertainty that determine which path we will take.”
Note
6. The Parliamentary Assembly must assess these concerns, by
analysing the potential misuses of AI systems, their impact on democracy,
human rights and the rule of law, and how these can be mitigated
or neutralised.
7. In preparing this report, the Committee on Political Affairs
and Democracy held different hearings throughout 2025:
- on 5 March, with the online
participation of Mr Daniel Innerarity, Chair in AI & Democracy
of the School of Transnational Governance – European University
Institute (who also further contributed with some additional considerations
on geopolitical implications of AI, drafted by him and other colleagues
of the European University Institute);
- on 24 June, with the participation of Mr Courtney Bowman,
Global Director of Privacy & Civil Liberties Engineering of
Palantir; Ms Audrey Herblin-Stoop, Vice-President and Head of Global
Public Affairs and Communication of Mistral AI (online); Ms Julie
Lavet, Head of Policy and Partnerships for Europe of OpenAI; and
Mr Ben Nimmo, Principal Threat Investigator of OpenAI (online);
- on 10 December, with the participation of Ms Francesca
Fanucci, Senior legal advisor of the European Center for Not-For-Profit
Law (ECNL), and representative of the Conference of International
NGOs of the Council of Europe at the Committee on Artificial Intelligence
of the Council of Europe.
8. The rapporteur wishes to thank all the above-mentioned participants
for their contributions, as well as the Committee members for their
comments and support in preparing this report.
2 Previous and current work of the Parliamentary
Assembly
9. The Council of Europe is addressing
the impact of AI technologies on human life in a comprehensive manner
and as a cross-cutting priority. The Assembly is an important contributor
to this work. In particular, it has a dedicated Sub-Committee on
Artificial Intelligence and Human Rights, and in October 2020 it
adopted a series of resolutions and recommendations on the matter
Note: among these, I was rapporteur for
the report on “Need for democratic governance of artificial intelligence”.
Note
10. In a common Appendix to these reports, the Assembly outlined
the ethical principles that it believes should be applied to AI
systems: transparency; justice and fairness; responsibility; safety
and security; privacy.
11. In
Opinion 303
(2024), the Assembly also welcomed the finalisation of the
Framework Convention on Artificial Intelligence and Human Rights,
Democracy and the Rule of Law, and stated that it would continue
to work on AI-related issues.
13. Furthermore, the Assembly co-organised a Parliamentary Conference
on Artificial Intelligence with the Parliament of the United Kingdom,
in London, on 15-16 December 2025. This Conference provided an opportunity
to exchange best practices and define parliamentary roles in AI
governance, and it yielded several relevant elements for this report.
Note
14. Finally, it is also worth mentioning that the Parliamentary
Assembly is currently working on a number of other relevant reports
on:
- “Digital transformation:
the role of the OECD in evaluating the impact of artificial intelligence
on the future of work”;
- “The need to modernise International Humanitarian Law”
(which will also take into account the use of AI in warfare);
- “Use of artificial intelligence by parliaments: risks
and opportunities”;
- “Safeguarding human rights in the artificial intelligence-driven
public sector”;
- “Safeguarding creativity and education in the age of generative
artificial intelligence”;
- “Artificial intelligence and gender equality: risks and
challenges”.
15. The ongoing work of the Assembly underscores the importance
it attaches to analysing the impact that AI is having on multiple
levels throughout European countries, with the aim of identifying
critical areas and defining solutions for this significant challenge.
The Assembly will continue to prioritise this matter and support efforts
to address the evolving implications of AI.
3 The
threats of artificial intelligence to democracy
16. The Council of Europe Framework
Convention on Artificial Intelligence and Human Rights, Democracy and
the Rule of Law (CETS No. 225) defines an “AI system” as “a machine-based
system that, for explicit or implicit objectives, infers, from the
input it receives, how to generate outputs such as predictions,
content, recommendations or decisions that may influence physical
or virtual environments”.
Note In other words, AI systems simulate
cognitive abilities and perform tasks that are normally associated
with humans: these include the analysis and treatment of data, problem-solving,
decision-making, and, most importantly, learning.
17. Current AI technologies are defined as “narrow” or “weak”
AI: they can perform specific, limited tasks, under certain instructions,
and often do so faster and more efficiently than humans. It is possible
that future developments might lead to “general” or “strong” AI,
which would be capable of applying what it has learned to different
contexts and matching human abilities. There is even the possibility
of “super” AI, which would be able to surpass human abilities and
potentially have its own “consciousness”.
18. The idea of a dystopian future in which democracy could be
replaced by 'algocracy', or rule by algorithms, through a conscious
super AI rather than elected representatives, is currently just
a hypothesis. However, given the pace of technological development
and its multiple implications for geopolitics, human rights, democracy,
the economy and the environment, it is crucial to monitor the very
real threats and disruptions that narrow AI is already causing to
democratic countries.
19. ChatGPT, a “generative” AI system and one of the most popular
AI chatbots
Note available
to the public today, shares the same view. When asked the question:
“Do you think you are a threat to democracy?”, the chatbot gave
the following answer:
“As an AI, my primary function is to assist
and provide information, not to influence decision-making or engage
in actions that could threaten democracy. However, the way AI systems
are used can certainly have implications for democratic processes.
AI can be misused by individuals or organisations to manipulate
public opinion, spread misinformation, or infringe on privacy. I’m
designed with ethical guidelines to promote positive interactions,
transparency, and respect for privacy. But AI systems like mine
are only as ethical and accountable as the policies, frameworks,
and people that govern them. The true threat to democracy would
arise if AI systems were exploited in ways that undermine trust,
equity, and fairness. That’s why it’s important for society to establish
clear regulations and safeguards around AI to prevent such risks.”
20. It is obviously important to consider how this response was
generated: AI chatbots are software capable of “interacting” with
human language, using sophisticated algorithms to process the request
made to them and produce results based on their access to very large
databases. In other words, the answer given by ChatGPT is based
on reports, articles, statements and publications of various kinds
available in its databases and on the Internet.
21. The way in which the algorithms are designed, however, can
greatly influence the results produced. For example, in the answer
provided above, ChatGPT states that it is “designed with ethical
guidelines” – but who defines these guidelines as ethical? And who
ensures that they are followed? As a result, the lack of transparency
of certain algorithms, as well as the fact that self-regulation
by the companies developing AI systems does not seem to be sufficient,
are additional issues that need to be assessed.
22. This is also reflected in public sentiments towards AI and
its governance: according to data from the Pew Research Centre,
a median of 34% of adults surveyed across 25 countries say they
are more concerned than excited about the increased use of AI
Note; according to
the Survey on Drivers of Trust in Public Institutions of the Organisation
for Economic Co-operation and Development (OECD), only 41% of respondents
in 30 countries thought it was likely that their national governments
would adequately regulate new technologies, such as AI and digital
applications, and help businesses and citizens use them responsibly;
Note and according to a research from
the Ada Lovelace Institute, a very large majority of people surveyed
in the UK believes that safety should take priority over speed,
and that the ability to ban AI on ethical grounds matters more than
competitive advantages.
Note
23. According to Stanford University, AI capability is now outpacing
the benchmarks designed to measure it, exceeding established human
performance. Furthermore, while benchmarking for responsible AI
(i.e. the set of practices and governance mechanisms designed to
ensure AI systems are safe, fair, and beneficial and that they perform
as intended) is increasing, it is not keeping up with AI advances
and deployments. Stanford University also reported that AI companies
have grown less transparent in 2025, increasing concerns about the
ways in which systems are developed, trained, tested and monitored.
Note
3.1 It
all starts from data
24. AI systems and tools rely heavily
on the ability to collect, store, and analyse large datasets, which
may include personal data. These large datasets are then used to
“train” many types of AI tools. Large-language models, for example,
are neural networks working on a probabilistic way, thus generating
their outputs based on patterns they identify in the datasets.
25. The misuse of AI systems by governments, private companies,
and other entities to access personal data without consent is one
of the most significant threats, due to its potential implications.
26. Personal data can be used for mass surveillance through public
video cameras, citizens’ mobile phone GPS locations, or their credit
card usage.
Note This is already widely implemented
in China, for risk scoring and predictive policing purposes, i.e.
to forecast possible future crimes and unrest, based on previous
police reports, surveillance data, and social media activity – with
a chilling effect on civil society activism.
Note
27. Furthermore, malicious actors can use personal data to create
fake identities for fraudulent purposes and scams. AI technology
can also be used to create deepfakes by cloning a person’s image
or voice to create fake images, audio files or videos (also known
as “synthetic media”), These can then be used to blackmail or harass
victims, or given high visibility and distributed by bots on large
social media platforms, to spread fake news. Among the worrying
recent examples, it is worth mentioning the hundreds of sexually
explicit deepfakes targeting women generated by Grok, the AI tool
integrated into the social media platform X.
Note
28. Datasets can be imbalanced and contain biases that are then
replicated in the outputs produced by AI systems. This can lead
to or reinforce discrimination, particularly against women and minorities,
when the systems are adopted for public policy-making purposes.
Note The algorithms used to analyse the
data and produce outputs can also be biased and lead to discrimination.
In this sense, AI systems and tools are seldom neutral; they often
inherit the biases of the societies in which they are built and
deployed.
29. Furthermore, individuals may be affected by potentially biased
algorithmic decision-making based on AI without their knowledge.
This undermines their right to an effective remedy, especially since
these systems often operate as “black boxes”, and responsibility
becomes almost impossible to attribute.
30. It is important to note that, according to some estimates,
the available pool of high-quality data for training AI models might
be fully utilised between 2026 and 2032, also raising concerns about
the possibility to scale up models and expand their capabilities.
Note
3.2 Disinformation
and the media
31. The World Economic Forum Global
Risks Report 2026 ranks “misinformation and disinformation” as the second
short-term most severe risk, just after “geoeconomic confrontation”.
Note
32. While misinformation and disinformation are not new phenomena,
the accelerating development and deployment of AI systems and tools
is certainly a major contributor to this risk. AI can be used to
fabricate fake news and disseminate them on a large scale with limited
resources and skills, and very quickly. As fake news travel faster
online than the truth,
Note and
considering the costs, resources and time needed to debunk them (often
without managing to effectively counter them), this represents yet
another major threat to democracies.
33. Malign foreign actors are using large-scale disinformation
campaigns, such as the Russian “Portal Kombat” network identified
by French authorities in February 2024.
Note By flooding the information
ecosystem with thousands of fake news stories, false statements
and deepfakes, these networks are poisoning the datasets on which
large language models are trained, therefore also influencing the
outputs that these models produce.
34. Similarly, “bot farms” and “troll farms” are increasingly
being used to disseminate fake news on social media platforms, by
creating fictitious profiles that are programmed to automatically
spread disinformation, hate speech and disruptive content. Platforms,
in turn, are not always efficient in moderating and removing fake content
– and in some cases, it is not in their interest to do so either:
their algorithms are structured in a way that prioritises the most
engaging content, allowing them to monetise it. Most of the times
this happens to be exactly the disruptive, divisive fake content
that should be removed. Therefore, platforms can exercise a very strong
gatekeeping power over political discourse, through their content
moderation capabilities.
35. On the other hand, information manipulation is also used to
distract individuals and generate distrust on institutions, activating
anti-democratic feelings and behaviours. Digital technologies and
generative AI are enabling the proliferation and fragmentation of
perceived realities, which poses a vital threat: democracy cannot
survive without some degree of shared reality.
Note
36. Furthermore, the rise of social media platforms and the more
recent deployment of AI tools are also affecting the business model
of traditional media. These platforms often feed on content without
any guardrails to protect intellectual property, which has serious
implications for the diversity and quality of information.
Note A concerning
estimate suggests that, in November 2024, the number of AI-generated
articles on the internet was larger than the number of articles
produced by humans.
Note
3.3 Political
persuasion and micro-targeting
37. AI technology appears to have
surpassed the capacity to persuade people better than humans do.
Note Digital technologies and AI systems
can therefore be used to influence citizens’ behaviour, particularly
with regard to their political attitudes, with increasing threats
appearing during elections. Foreign information, manipulation and
interference (FIMI) pose a significant threat to European democracies.
38. Recent examples include the first round of the Romanian presidential
elections on 24 November 2024, which were annulled by the Constitutional
Court due to sophisticated digital manipulation techniques reportedly deployed
on large social media platforms (in particular TikTok), most probably
by a foreign power
Note; and the Moldovan
parliamentary elections held on 28 September 2025, during which
Russia undertook a massive disinformation campaign, spending the
equivalent of over 1% of Moldova’s GDP in the effort.
Note
39. AI systems and tools can be used to micro-target specific
groups of citizens, for marketing purposes, as well as to influence
their political opinions and choices. This can involve reducing
their access to certain sources of information while magnifying
their exposure to others, and it is achieved through the psychological profiling
of users, who then receive highly tailored messages and content.
Paid influence may also be incorporated into AI-generated outputs,
in a way that is indistinguishable from neutral content.
40. Micro-targeting contributes to the creation of echo chambers,
in which citizens are not exposed to a variety of opinions and are
unable to engage in a political dialogue with those who think differently.
Instead, they are passive recipients of information and ideas that
reinforce their own beliefs.
41. Chatbots, in particular, can be highly effective in political
persuasion. A recent study in Canada and the US showed how they
could be used to persuade people to change their vote orientation.
Note Similarly, ahead of the parliamentary
elections held on 29 October 2025, the Dutch data protection authority
warned that AI chatbots were unreliable and clearly biased when
offering voting advice.
Note
42. In an increasing number of countries, “softfakes” are also
being used: these are synthetic images created to make political
candidates more appealing. While they are often created by the candidates themselves
(or their teams), their use in electoral campaigns still raises
ethical concerns.
Note
43. Finally, AI can also be used to censor specific content. The
Chinese AI chatbot DeepSeek (which has an overall performance comparable
to that of ChatGPT and other US-made chatbots, but costs significantly less)
Note, is censoring or biasing answers
to questions related to the Chinese government. This clearly demonstrates
how these tools can be used to influence users.
Note
3.4 Other
threats
44. The threats described above
can have an additional detrimental effect on democracy, through
the increased deployment of AI systems and tools, as they increase
people’s distrust of the media, public institutions and the integrity
of democratic processes. Not only is there a risk of people being
deceived by false content, but they may also become more cynical.
News fatigue might appear, and people may be led to rely on social
media platforms rather than traditional media (already a reality
for most young Europeans)
Note, or not believing any source of
information altogether. This further fuels polarisation and social
unrest, thus contributing to the backsliding of democracy.
45. An additional threat is posed by AI “hallucinations”, whereby
AI generates outputs containing false or misleading information.
These can reinforce discrimination and spread disinformation, further
polluting datasets and the information ecosystem. They can also
lead to defamatory results for individuals: in a recent case, a
Norwegian man asked ChatGPT for information about himself, and received
the false claim that he had murdered his two sons in reply.
Note
46. Furthermore, there is a risk that over-reliance on AI tools
could reduce people’s ability to think critically and innovate,
setting lower standards and limiting the diversity of perspectives.
Citizens might use AI to address questions to their elected representatives,
who might also use AI to provide their answers. Similarly, students
might use AI to produce their work, and teachers might then use
AI to correct it.
47. If incompetent AI is given decision-making responsibilities,
on the basis of unjustified forms of techno-optimism or misjudgements
about AI’s real capabilities, there is an additional risk of “incompetence
elevation”, which could lead to catastrophic results.
48. Using AI tools can also lead to loneliness and the erosion
of social skills, particularly when users anthropomorphise the tool
and use it as a synthetic substitute for real-life relationships,
including romantic ones.
Note
49. Ownership of datasets, coupled with control over the most
powerful AI systems and computing technology, can lead to dangerous
concentrations of power in the hands of private tech corporations,
especially in the absence of proper oversight.
Note
50. In addition, different levels of access to technology and
to the skills required to use it fully can exacerbate existing inequalities
between citizens. Women, older people, and marginalised groups with
lower access to education may be the most vulnerable. While it is
too early to foresee the net effects of AI on the labour market, it
is possible that certain groups might be left behind in the digital
transition: those who will lose their jobs due to the increased
automatisation of tasks might become more alienated and inclined
towards political extremism.
51. Finally, the increasing demand for AI services, and the related
need for ever-larger data storage and computational capabilities,
can seriously impact energy production and the environment, particularly
with regard to CO2 emissions and water consumption.
4 How
can artificial intelligence strengthen democracy?
52. AI should not be demonised,
however, as it can also strengthen democratic processes, if properly developed,
deployed, and monitored. Although there are risks, AI also offers
opportunities. With the right governance, standards and rules in
place, these opportunities could offset the threats, converting
AI in a formidable vehicle to strengthen democratic institutions
and processes, making them more resilient and ready for the challenges
of the future. There is still time to make this change happen.
53. The most obvious application of AI systems in politics is
to encourage public participation. In particular, AI could provide
citizens with access to specific information; analyse and explain
otherwise technically complicated policies; mediate and summarise
discussions, identify patterns and common positions among large
groups of people involved in consultations and deliberative processes;
and facilitate their interaction with their elected representatives
and public authorities.
Note In this sense, AI could in a near
future be used to facilitate international consultations and identify
shared solutions to global issues.
Note
54. Similarly, AI could also be used by elected representatives
and public authorities to gather feedback, proposals, and information
about the needs and desires of citizens in their community. This
would support policy- and decision-making processes, making them
more focused and effective.
55. Furthermore, AI tools can have a positive impact on inclusiveness,
as they could provide people from all backgrounds, and especially
disadvantaged groups, with better access to public services, education
and job opportunities, thereby levelling the playing field and eliminating
barriers.
56. AI could therefore encourage citizens to participate more
in the public sphere, bringing them closer to the politicians who
take decisions that affect their lives. This could, in turn, reverse
the trend of abstention, as citizens would regain trust in institutions
by feeling that their voices are heard and considered.
57. AI can also be used to strengthen the protection of human
rights (e.g. by countering human trafficking)
Note or to automate and speed up certain
public administration procedures, thereby increasing productivity, responsiveness,
accountability,
Note and
in turn improving citizens’ satisfaction with the work and performance
of their local and national governments.
58. Several parliaments in Europe are already piloting AI tools
to improve their internal work, including the Assembly. In September
2025, Albania launched Diella, an “AI minister” responsible for
public procurement, with the aim of making the process transparent
and incorruptible.
Note
59. Finally, AI technology could be used to monitor and detect
attempts to influence citizens with the help of AI itself, e.g.
by identifying AI-generated content, and in general, it could be
used to strengthen democratic oversight of the development, testing
and application of AI systems and tools.
5 Geopolitical
implications
60. The race to develop and control
AI technology, its infrastructure and the relevant resources will determine
the course of geopolitical relations at global level in the years
to come.
61. The first implication concerns the “digital sovereignty” of
countries. Ownership and control of datasets and data centres, cloud
computing capabilities (including quantum computing), human skills,
AI research and development hubs, and sufficient energy grids will
become increasingly valuable assets for national security and state
relations: countries with this control will be able to manipulate
and selectively “shut down” AI services used by countries lacking
it, as a foreign policy instrument.
62. The second implication refers to the clash between competing
visions of society and political regimes, which also influence different
“models” for AI development. The United States is still leading
the way in AI research and development, thanks to their model characterised
by an innovation ecosystem grounded on free markets and competition,
and in which their tech corporate giants operate. China, on the
other hand, is catching up through its single-party, state-centralised
planning, and it is shaping a model through which the strong presence
of the central government in citizens’ lives is reflected in the
development and deployment of AI systems and tools.
63. In turn, the EU is criticised of lagging behind because of
its focus on regulation – based on the fallacious argument that
regulation is an impediment to innovation. It is important to note
that this is a false dichotomy: appropriate regulation can facilitate
technological development, by providing innovators with a safe and predictable
environment. Europe’s real challenges may lie in its fragmented
capital markets, the lack of a single digital market, punitive bankruptcy
laws and higher risk aversion, which all contribute to having less
incentives for the creation of an innovation ecosystem, especially
for what concerns start-ups.
Note
64. The “EU model”, however, particularly emphasises digital rights
and an effort to project normative leadership globally, in order
to shape international regulatory norms (the so-called “Brussels
effect”).
65. The relevance of AI as a geopolitical asset lies therefore
in its ability to reshape power relations, international norms and
the very existence of democracy. It also represents a domain of
ideological competition, with authoritarian regimes and democratic
countries competing for the prevalence of their AI systems, which
will in turn embed different sets of values.
66. In addition, the concentration of control over AI technologies
in the hands of a small number of tech giants, which have greater
financial resources than many States, is concerning for two reasons.
Firstly, the imbalance of power between private companies and countries
with fewer resources is significant and has geopolitical implications.
Secondly, the transnational nature of these companies makes it more
difficult to ensure accountability.
67. One recent example is the case of Anthropic’s cyber security
model Mythos, whose release was limited to a number of tech companies
and organisations: while the model seems to be outperforming humans
in identifying cyber vulnerabilities, companies and even government
bodies which are not granted access find themselves at a disadvantage.
An additional concern is raised by the fact that such technology,
in the hands of a malign actor, can also become a threat.
68. Furthermore, some observers are indicating that the rapid
commercialisation of products to ensure a fast return on current
investments poses a major risk for 2026. This could lead some companies
to deploy AI tools that have not yet been properly tested.
Note
69. Finally, it should be noted that a country capable of achieving
general or super AI would pose an unprecedented threat to its rivals.
This may eventually lead to “Mutual Assured AI Malfunction” deterrence frameworks
(similar to the Mutual Assured Destruction deterrence framework
ensuring stability during the nuclear era).
Note Even worse, super AI may be impossible
to control, with consequences that are even difficult to imagine
at this stage. One possible way to handle this could be to follow
the model of Nuclear Treaties.
Note
70. In September 2025, the AI Red Lines Initiative was launched,
warning that it could become increasingly difficult to exert meaningful
human control over unchecked AI systems in the coming years, and
urging governments to “reach an international agreement on red lines
for AI – ensuring they are operational, with robust enforcement
mechanisms – by the end of 2026”.
Note
71. Similarly, a Statement on Superintelligence was published
in October 2025, collecting signatures from hundreds of public figures,
calling “for a prohibition on the development of superintelligence,
not lifted before there is broad scientific consensus that it will
be done safely and controllably, and strong public buy-in”.
Note
6 Current
international efforts
72. Although technological advancements
have a global impact, the regulations governing them seldom transcend
national borders. This highlights the need to strengthen international,
multi-stakeholder cooperation.
6.1 Council
of Europe
73. The Council of Europe is contributing
to this international effort in different ways. Its Framework Convention
on Artificial Intelligence and Human Rights, Democracy and the Rule
of Law was opened for signature in Vilnius on 5 September 2024,
and represents the first-ever international legally binding treaty
in the field of AI.
Note
74. Article 4 of the Framework Convention states that the signatory
Parties “shall adopt or maintain measures to ensure that the activities
within the lifecycle of artificial intelligence systems are consistent
with obligations to protect human rights”. Furthermore, Article
5 provides that “[e]ach party shall adopt or maintain measures that
seek to ensure that artificial intelligence systems are not used
to undermine the integrity, independence and effectiveness of democratic
institutions and processes”, and “that seek to protect its democratic
processes in the context of activities within the lifecycle of artificial
intelligence systems, including individuals’ fair access to and
participation in public debate, as well as their ability to freely
form opinions”.
75. Furthermore, in November 2024, the Committee on Artificial
Intelligence (CAI) established by the Committee of Ministers of
the Council of Europe adopted the Methodology for the risks and
impacts of artificial intelligence systems from the perspective
of human rights, democracy and the rule of law (HUDERIA Methodology)
Note, a non-legally binding guidance
document intended to be used by public and private actors. As of
1st January 2026, the work of CAI is
carried on by the newly established Steering Committee for New and Emerging
Digital Technologies (CDNET) as the CAI’s successor committee.
76. In November 2025, the Steering Committee for Human Rights
adopted the Handbook on Human Rights and Artificial Intelligence,
designed to support government officials and policymakers in Council
of Europe member States in applying the European Convention on Human
Rights (ETS No. 5), the European Social Charter (revised) (ETS No.
163) and other human rights standards to the use of AI.
Note
77. In April 2026, the Committee of Ministers of the Council of
Europe adopted Recommendation CM/Rec(2026)4 on online safety and
empowerment of users and content creators, calling for greater transparency, accountability
and oversight regarding how platforms hosting user-generated content
design their interfaces and algorithms, and assess and manage the
risks they pose to users’ human rights and democratic processes.
Note
78. Overall, AI is being treated as a transversal priority matter
by the Organisation; other relevant work includes the Guidance note
on the implications of generative artificial intelligence for freedom
of expression, adopted in December 2025 by the Council of Europe
Steering Committee on Media and Information Society.
Note
79. The Committee of Experts on Foreign Information Manipulation
and Interference (PC-FIMI), established as a sub-committee under
the European Committee on Crime Problems (CDPC), has prepared a
Feasibility study on the possible elaboration of a legal instrument
on foreign information manipulation and interference (FIMI), including
disinformation, presented at the Ministerial session of the Committee
of Ministers in Chişinău (Republic of Moldova), on 14-15 May 2026.
Note
80. Furthermore, the Steering Committee on Democracy is currently
finalising a draft Study on the advantages and risks of the use
of generative artificial intelligence in public debate relating
to democratic processes and on AI literacy for democratic life.
81. The Council of Europe has been paying close attention to the
participation of young people and their organisations in AI governance.
In April 2025, it held a consultative meeting which resulted in
a Roadmap on Artificial Intelligence, Youth Policy and Youth Work.
Note
82. The Council of Europe also launched in 2025 the New Democratic
Pact for Europe, to boost solutions that work and create new responses
to democratic backsliding in the continent.
Note AI will be at the core of this effort,
particularly for what concerns its Pillar 3 “Innovating for Democracy”,
which aims at harnessing digital technologies for good – including
AI – while combating their misuse and harmful impacts. This report
and the other relevant ones currently being prepared by the Assembly
will be a crucial contribution.
6.2 United
Nations
83. The United Nations (UN) have
integrated a Global Digital Compact in the Pact for the Future adopted
in September 2024,
Note in which member States had
set as an objective to close all digital divides and enhance international
governance of AI for the benefit of humanity.
84. Through the Compact, member States also decided to establish
a multidisciplinary Independent International Scientific Panel on
Artificial Intelligence
Note and initiate a Global Dialogue on
Artificial Intelligence Governance. The first session of the UN
Global Dialogue will be held in July 2026.
Note
85. Furthermore, the UN Secretary General warned in a public message
against the dangers of unchecked AI for democracy, and called for
AI to serve humanity equitably and safely.
Note
86. Different UN Agencies and entities are working on AI from
various angles. UNESCO
Note is tackling issues related to ethics
of AI, AI in education, gender equality and capacity building for
governments and judiciary. It also produced a dedicated report on
“Artificial intelligence and Democracy”
Note. UNWOMEN also provides a gender
focus on AI, and last year launched an AI School which allows participants
to learn, build and govern safe, gender-inclusive AI.
Note
6.3 G7
and the Organisation for Economic Co-operation and Development (OECD)
87. The G7 is paying close attention
to AI developments; in particular, through the G7 Hiroshima AI Process, it
developed in 2023 the Guiding Principles and a Code of Conduct for
AI actors.
Note
88. The OECD is also intensively working on AI-related issues;
the scope of its work, especially for what concerns AI’s impact
on the labour market, will be thoroughly revised through a dedicated
report currently being prepared by the Assembly. It is however worth
to mention that the Council of the OECD adopted in 2019 (and amended
in 2024) a Recommendation on Artificial Intelligence.
Note Furthermore, the OECD cooperates
with the Global Partnership on AI.
Note
6.4 European
Union (EU)
89. The EU has undertaken several
relevant initiatives. In 2025 it adopted an AI Continent Action
Plan, to develop trustworthy AI technologies to enhance Europe’s
competitiveness while safeguarding and advancing democratic values.
Note
90. In 2024 it adopted Regulation (EU) 2024/1689 laying down harmonised
rules on artificial intelligence (AI Act)
Note, which sets out a risk-based approach
for AI developers and deployers regarding specific uses of AI, defining
4 level of risks for AI systems: unacceptable, high, limited and
minimal. In particular, the AI Act prohibits AI systems aimed at
establishing social ratings, profiling individuals to predict their
risk committing a crime, or exploiting peoples’ vulnerabilities
to distort their behaviour.
91. Furthermore, the EU adopted in 2022 Regulation (EU) 2022/2065
on a Single Market For Digital Services (Digital Services Act),
introducing rules for online services used by European citizens
in their everyday life, including marketplaces, social media networks,
app stores, and online travel and accommodation platforms.
Note
92. Finally, the EU also launched the European Democracy Shield
initiative in November 2025, containing a series of concrete measures
to empower, protect, and promote strong and resilient democracies
across the EU.
Note
6.5 Other
initiatives
93. Other relevant initiatives
include the AI Safety Summit, held in Bletchley (UK) in November
2023;
Note the Paris AI Action Summit held
in February 2025;
Note and the AI Impact Summit held in
New Delhi in February 2026.
Note
94. In November 2024, the AI safety institutes and government-mandated
offices that facilitate AI safety and evaluation from the European
Commission, France, the United Kingdom as well as Canada, Japan
and the United States, among other countries, launched the International
Network of AI Safety Institutes. The aim is to facilitate a common
technical understanding of AI safety risks and mitigations, and
to encourage a general understanding of and approach to AI safety
globally.
Note
95. In addition, the International Organization for Standardization
(ISO) created in 2017 the standardization subcommittee “ISO/IEC
JTC1/ SC 42 Artificial Intelligence”, focusing on the entire AI
ecosystem, to enable mass deployment and adoption of AI in a large
number of fields.
Note
96. Finally, it is also worth mentioning the mapping of public-interest
initiatives leveraging AI to support democratic governance online,
prepared by the civic tech organisation Make.org.
Note
7 Conclusions
97. In 1955, Isaac Asimov published
the short story “Franchise”, which imagined a future in which the
United States has become an “electronic democracy”. In this future,
a supercomputer selects a single individual to represent the entire
electorate and respond to a set of questions. The machine would
then use these answers to determine the election results. “The machines
grew bigger and they could tell how the election would go from fewer
and fewer votes. Then, at last, they built Multivac and it can tell
from just one voter”. Although this troubling scenario currently
seems to be confined to the realms of science fiction, it is worth
noting that in September 2025 a new political party in Japan identified
AI as its leader.
Note
98. Meanwhile, the most recent developments in digital technology
and AI pose very real threats to democratic institutions and processes,
demanding immediate action from Council of Europe member and observer
States.
99. Although the digital transition may bring many benefits to
humankind, the risks to democracy cannot be underestimated. The
Council of Europe is a pioneer: its Framework Convention on Artificial
Intelligence and Human Rights, Democracy and the Rule of Law is
the first-ever legally binding treaty in this field.
100. This is, however, not enough. Member and observer States must
adopt measures to ensure that democratic oversight is maintained
throughout the entire lifecycle of all AI systems, from development
to deployment, to safeguard human dignity, transparency, and accountability.
This must be achieved in collaboration with all the relevant stakeholders,
starting from the private actors operating in the field of AI, to ensure
that AI systems are truly human-centred AI and empower humans without
replacing them.
101. Initiatives to counter foreign interference and strengthen
resilience against disinformation should be prioritised.
102. AI developers, providers, and social media platforms should
be held accountable for any harm caused by their services, and victims
should have access to redress measures.
103. It is equally important to dedicate sufficient resources to
AI and media literacy, and to involve all relevant stakeholders
in developing new standards and regulations.
104. Member States should seek to define clear frameworks for their
relations with big tech corporations, while also taking steps to
strengthen European digital sovereignty.
105. A multilateral body dedicated exclusively to overseeing AI
technologies could be established, to define a common language,
standards and a regulatory framework, involving all interested stakeholders.
106. Ultimately, it is important to recognise that AI technologies
can also be a powerful tool to strengthen democratic security in
Europe. Member States must be prepared to seize this opportunity
and enjoy the resulting benefits, while ensuring that proper governance
frameworks are in place.