As artificial intelligence (AI) becomes part of everyday life, one question matters more than ever: do people actually trust it? A new study from the CERTAIN project looks beyond technology to explore social acceptance – how people understand and evaluate AI in real-world contexts. Based on discussions across Armenia, Estonia, Kazakhstan, Slovenia, and Spain, one message stands out:
AI is not judged by what it can do, but by how – and by whom – it is used.
AI Is About Institutions, Not Just Technology
Participants did not see AI as a neutral tool. Instead, they viewed it through the lens of institutions, power, and real-world impact – from public authorities to corporate actors and even global competition.
Across all discussions, the same questions emerged: who is using AI? For what purpose?Under what rules?
This points to a key insight: trust in AI depends on trust in the institutions deploying it.
Acceptance Is Conditional
Acceptance of AI is never absolute, but is always conditional.
Participants consistently said AI is acceptable only if humans remain in control, decisions can be explained and challenged, responsibility is clear and data is handled responsibly. Without these safeguards, even advanced systems are met with scepticism. Acceptance is not about capability – it is about conditions.
Human Oversight Is Essential
Across all countries, participants drew a clear line: AI may support decisions, but should not replace human judgment. This was especially important in areas affecting people’s lives, such as public services, education, and healthcare. Human involvement was seen as essential for responsibility, context, and the ability to correct mistakes. Concerns about AI were less about the technology itself, and more about what happens when human responsibility becomes unclear.
Data Protection and Fairness Matter Most
Concerns about data were widespread and practical. Participants worried about:
– Losing control over personal information
– Unclear data sharing or reuse
– Potential misuse by institutions or third parties
A key concern was that once data enters AI systems, control may be lost over time.
Fairness was equally central. Participants questioned whether AI systems:
– Treat people consistently
– Reflect individual circumstances
– Allow decisions to be challenged
Fairness was closely tied to transparency and contestability. Systems that are rigid or opaque were widely seen as unfair.
Shared Expectations, Different Concerns
Across countries, expectations were strikingly similar: human oversight, accountability, data protection, and fairness.
Some emphasised regulatory gaps, others misuse or loss of control, and in several cases, concerns extended to data sovereignty and geopolitical influence.
What This Means for AI Governance
The message is clear: technical performance alone does not build trust.
Trust depends on visible safeguards:
– Human oversight
– Clear responsibility
– The ability to challenge decisions
AI systems must be understandable, accountable, and aligned with societal expectations.
A Social Foundation for Trustworthy AI
The findings highlight a simple but critical point: AI governance must be grounded not only in law and technology, but in how people actually experience and evaluate AI. By bringing these perspectives into focus, the CERTAIN project helps shape AI systems that are not just innovative, but also trusted, fair, and socially accepted.