Table of contents:
A new survey, conducted by AtlasIntel for Edge Institute, shows that more than 8 in 10 Romanians use artificial-intelligence tools and, moreover, that a majority of Romanians (over 75%) use these tools every week. When it comes to the state, acceptance of these tools varies: Romanians support them for administrative and repetitive tasks and reject them for decisions involving health and rights.
The norm: adoption
A year after the first Call for Romania’s Digital Transformation and a few weeks after the second Call, in which we wrote that Romanian citizens use artificial intelligence daily while the Romanian state looks on passively, the new Edge Institute survey cements this observation. Conducted on a sample of 2,002 respondents between 17 and 21 June 2026, with a margin of error of 2% and a 95% confidence level, the survey measures, among other things, how Romanians use artificial intelligence in their lives and the extent to which they would accept institutions using it in their dealings with citizens.
Just over 83% of Romanians use artificial-intelligence tools in at least one context, while 16.9% say they do not use them at all. Most use is personal, reported by 77.9% of respondents, while 15.4% use such tools at work and 9.3% at school or university. The fact that professional and educational use lags behind personal use shows, on the one hand, that Romanians reached artificial intelligence ahead of any public policy, any training programme and any institutional guidance that would integrate AI into work or learning. On the other hand, this pattern may of course also reflect a degree of stigma around admitting to using AI at work or school.
More than eight in ten Romanians use AI
Do you use artificial intelligence-tools?
% of total, multiple response
Among those who use artificial intelligence, 50.7% do so a few times a week and 26.2% daily, which means that almost 77% of users interact with these tools at least weekly. At the same time, skill levels remain low: when assessing their own level, 48.1% of users place themselves at a basic level and 40.8% at an intermediate level, while 11.1% consider themselves advanced users, a sign that use is running ahead of the acquisition of anything more than minimal skills.
How often they use AI
Among those who use artificial intelligence, %
Among those aged 18–29, 97% use artificial intelligence, and the share falls to 68.5% in the 50–59 group and to 29.3% among those over 70; by education, it rises from 80% among those without higher education to 93.3% among higher-education graduates.
Who uses artificial intelligence
Benefits and risks, both acknowledged and accepted
Romanians weigh the gains and the costs of artificial intelligence at the same time. The benefit most widely recognised is saving time, which 65.4% of respondents associate with artificial intelligence, producing a net agreement of almost 48 points. At the same time, 55.5% agree that artificial intelligence will make some jobs disappear, and 51.8% believe it creates risks for personal data and security.
What Romanians think about AI
Net agreement, percentage points(% agree minus % disagree)
Fear for jobs, felt more at the level of society
The idea that artificial intelligence spreads more disinformation than people do produces a negative balance of 6.1 percentage points, while the idea that artificial intelligence helps people learn and develop their skills is endorsed by 42% of respondents but meets rejection from more than 30%.
The conclusion is that, when it comes to artificial intelligence, Romanians need information, clear rules and context: the public already assesses the technology in a relatively balanced way, recognises the gains, accepts the risks and thus calls on the state to act.
Although 55.5% of respondents believe that artificial intelligence will eliminate jobs across society, only 21.9% say they are worried or very worried that their own job, current or future, could be affected in the next five years, while 51.1% are slightly or not at all worried. This gap between the risk perceived at the level of society and the risk felt personally appears in many contexts, but in the Romanian case it carries a particular significance, which we described in our article on Romania in the age of artificial intelligence and in our report on Automation in Romania.
We have argued that the danger lies less in layoffs and more in a decline in productivity and income among the groups that use the technology without benefiting from the productivity gains it brings. In this sense, personal optimism may reflect confidence in one’s own ability to adapt, but it may also be an underestimation of the transformation to come, particularly among those with limited access to training; one more reason for the Romanian state to prepare (re)skilling programmes in good time.
Artificial intelligence in state institutions
Asked how comfortable they would feel about state institutions using artificial intelligence in various situations, Romanians distinguish routine administrative tasks from decisions that could more deeply affect their lives and rights. Comfort is lowest when artificial intelligence would make a medical diagnosis, where 69.7% say they are uncomfortable and 15.5% comfortable, and it remains low when it would decide on social benefits, with 57.4% uncomfortable against 26.6% comfortable.
The ratio reverses for administrative tasks: when artificial intelligence would answer citizens’ questions in the form of a chatbot, opinion is balanced, with 39.3% comfortable and 42% uncomfortable; and when it would triage requests and case files or evaluate the performance of civil servants, those who are comfortable outnumber those with reservations, 38% against 36.1% in the first case, and 42.2% against 37.2% in the second. The public’s message can also be translated into governance principles: artificial intelligence is accepted where it speeds up administrative work and relieves civil servants of repetitive tasks, and rejected where it would replace human judgement in decisions about health or rights.
Where citizens accept AI in state institutions
How comfortable would you be for the state to use AI to...? %
This distinction is also found in the experience of other countries. In the United Kingdom, for example, the government launched in 2025 a package of artificial-intelligence tools for civil servants, called Humphrey, which includes tools for summarising public consultations, transcribing meetings and preparing briefings, and it published an official playbook for using artificial intelligence in government. In Estonia, the state is building Bürokratt, a national virtual assistant through which citizens can access public services via a single conversational channel, with a large-language-model component funded from European funds. Denmark illustrates the reverse: its automated welfare-fraud detection system, which uses more than 60 artificial-intelligence and machine-learning models to flag beneficiaries, was examined by Amnesty International in a November 2024 report that describes extensive surveillance and a risk of discrimination against people with disabilities, people on low incomes, migrants and minorities, to the point of potentially qualifying as a social-scoring system banned by the EU AI Act. In short, the areas where Romanians show the greatest comfort are already being implemented in European states, and Romania has a duty to learn and draw inspiration from such pioneering countries.
The demand exists, the supply lags
The findings on artificial intelligence also fit within the wider context of digitalisation: Romanians view digitalisation favourably, since 75.4% consider it a good or very good thing, and interaction with digital public services has become the norm (77.6% used such services in the past year). Where services exist and are used, they satisfy citizens: 61.9% of users say they are satisfied or very satisfied with their online interactions with institutions, which shows that the obstacle lies with what the state offers, and less with citizens’ willingness, all the more so given that the services citizens have interacted with belong to a narrow range, such as paying taxes and duties and booking appointments online. All these figures are also stable across earlier surveys (for example, a 73% positive attitude towards digitalisation in November 2025 and 78% in March 2026), which further validates these perceptions.
Even though demand for digital services is high, 81.6% of Romanians consider the pace of the state’s digitalisation to be slow or very slow, and 48.3% judge Romania to be little or not at all digitalised. Likewise, these beliefs are already entrenched, having been measured in our earlier surveys too: more than 84% considered the pace of digitalisation slow or very slow in November, and more than 86% in March.
As for citizens’ priorities, health comes first, with 52.5% calling for online appointment booking at state hospitals and a medical record accessible online, and 41.7% calling for prescriptions, certificates and referrals online, followed by online vehicle registration at 41.2% and a single application holding all personal documents at 40.5%. This priority also overlaps with the European Digital Identity Wallet (EUDI Wallet) project, which we addressed in the Call and which Romania is obliged to deliver by the end of 2026.
High demand, slow delivery
Five priorities, drawn from Romanians' own voices
The survey also shows to whom this demand is addressed: asked who should be responsible for Romania’s digitalisation, 60% of respondents point to the Government and the prime minister, ahead of Parliament at 40.4% and of universities, political parties or the European Union. Similar figures were recorded in the March survey (40% Parliament, 70% Government/prime minister) and in the November 2025 survey (40% Parliament, 75% Government/prime minister). We thus have a clear picture of a reality in which citizens place responsibility exactly where we have argued capacity must be built: at the Centre of Government, around a clear, unified and accountable mandate and implementation structure.
The first priority is governance, because the public places responsibility at the Centre of Government. Confirming a Deputy Prime Minister with a dedicated mandate for digital transformation, and appointing the president of the Authority for the Digitalisation of Romania as the Government CIO with authority over the ministries, are the preconditions that any intervention in the area of artificial intelligence assumes. Within this mandate, a mandatory guide for using artificial intelligence in public administration should be adopted by Government Decision and with the involvement of the Presidential Administration as a pole of expertise, so that the uncoordinated use of commercial tools, what we have called Shadow AI, is replaced by a safe and measurable practice.
The second priority is to start where citizens are already keen to see implementation. Back-office applications, from summarising documents and transcribing meetings to triaging case files, are the area where the public places trust and where the British Humphrey model shows that efficiency gains can be achieved; and a conversational assistant for citizens, built along the lines of Estonia’s Bürokratt, could meet this citizen demand.
The third priority is to draw limits where the public (and the European Union and human rights) demand them. For example, decisions on social benefits and medical diagnosis, alongside public safety, should be kept under human review: alignment with the EU AI Act, which places such uses in the high-risk / unacceptable category, provides the framework, while the national implementing legislation, together with making the supervisory authority operational, remains a task the next government must treat as a priority.
The fourth priority concerns skills: a public that uses artificial intelligence but describes itself as an “advanced user” in only 11.1% of cases needs training, both within the administration, through programmes coordinated by the National Agency of Civil Servants and the National Institute of Administration, and at the level of citizens, through the involvement of local authorities, so that the gap between young and old and between urban and rural narrows. The split in opinion about the technology’s educational value supports investment in digital literacy, since perceptions can be expected to change when exposure is accompanied by the acquisition of skills.
How well AI users think they master it
Self-reported skill level among AI users, %
The final priority is transparency and measurement through data collection, because a state that uses artificial intelligence with citizens’ support has much to train, measure, build and protect. Pilot programmes in health and education, reported publicly and with efficiency indicators; a public-procurement framework for artificial-intelligence products and services integrated into national procedures; and a regulated testing environment are the instruments through which the conditional acceptance expressed by the public becomes earned trust.
Conclusion
The survey on the digitalisation of public services confirms what we described in our latest Call: the Romanian citizen has adopted artificial intelligence, uses it weekly, and knows what they want it for and where they set its limits.
Romanians accept the technology where it saves time and relieves the administration, and reject it where it would make medical or social decisions. The state now has the mandate, one the data places squarely before it, to deliver in the areas citizens view with openness and to protect in the areas where caution or prohibition is needed.
As for the production of knowledge, these nationally representative data provide a particularly relevant starting point for continuing research, so that the use of AI in Romania can be explored in greater depth.
Sources and references
- AtlasIntel for Edge Institute, “The Digitalisation of Public Services in Romania III”, survey conducted 17–21 June 2026, sample of 2,002 respondents, margin of error two percentage points, 95% confidence level.
- Edge Institute, The Second Call for Romania’s Digital Transformation (2026)
- Edge Institute, Romania in the age of AI: overexposed, unprepared, absent
- Edge Institute, Human dignity through the governance of Artificial Intelligence
- OECD, Governing with Artificial Intelligence (2025)
- UK, AI experiments see “Humphrey” help townhalls cut costs and improve services (2025)
- Government Digital Service, Launching the Artificial Intelligence Playbook for the UK Government (2025)
- European Commission, #Bürokratt programme and national virtual assistant platform and ecosystem
- Amnesty International, Denmark: AI-powered welfare system fuels mass surveillance (November 2024)
The current wave 2.002 respondents, with data collected between 17–21 June 2026. The comparative analysis integrates previous waves. All three waves use the same Atlas methodology — random digital recruitment — with a margin of error of ±2 p.p. at the 95% confidence level.
ⓘWhere the wording of the questions or the list of options has changed between waves, the value comparison is indicative.