Trust is one of those concepts we use every day without thinking too much about it. We trust people, institutions, technologies, and countless invisible systems that make modern life possible. Yet when we try to explain what trust actually is, the concept quickly becomes more complex.
The sociologist and philosopher Niklas Luhmann famously wrote1 that without trust we would not even be able to get out of bed in the morning. Without it, an undefined anxiety and paralyzing uncertainty would overwhelm us. Trust allows us to act despite the complexity of the world.
This idea echoes an even earlier insight from John Locke2, who described trust as the vinculum societatis — the bond that holds society together. Every day we rely on the cooperation and reliability of others: when we drive in traffic, we trust other drivers to follow the rules; when we eat food, we trust that many different actors in the supply chain ensured its safety.
In other words, we can “get up in the morning” because we assume that others will behave in reasonably predictable and cooperative ways.
But how exactly does trust work?
Trust as a Mental State, a Decision, and a Behavior
The socio-cognitive model of trust proposed by Cristiano Castelfranchi and Rino Falcone (2010)3 helps break trust down into a series of steps.
Trust does not appear all at once. It develops through a process that moves from evaluation to decision and finally to action.
First, trust exists as a mental attitude. We form an opinion about whether another person, organization, or system might be reliable in a particular situation.
Second, trust becomes a decision. We evaluate the situation and decide whether relying on that person or system is reasonable.
Finally, trust turns into behavior. We actually act on that decision — for example by following advice, sharing information, or relying on someone to complete a task.
Trust, therefore, is not only a feeling. It is a cognitive process that eventually leads to concrete actions.
The Two Core Ingredients of Trust: Goals and Beliefs
According to the socio-cognitive model, trust is based on two fundamental ingredients: goals and beliefs.
In its simplest form, trust arises when a person has a goal they want to achieve, but reaching that goal requires relying on someone or something else.
The decision to rely on that person or system is shaped by beliefs about whether the goal is likely to be achieved.
Among the most important beliefs are:
- Competence – whether the other party has the skills or abilities required to help achieve the goal.
- Willingness – whether they intend or are motivated to do so.
- Unharmfulness – whether they are unlikely to interfere negatively with achieving the goal.
- Dependence – whether the goal can be reached independently or only through their involvement.
- Context – whether the surrounding circumstances support or hinder success.
These elements interact in the mind of the person deciding whether to trust. The decision is also influenced by the risk involved, the value of the goal, and the sources from which the beliefs originate.
Sources of trust can include direct experience, reputation, recommendations from others, or personal reasoning. Each of these sources is evaluated in terms of relevance, certainty, and credibility.
Trust Offline and Online
In everyday offline interactions, trust is often built gradually through experience, repeated interactions, and social cues. Facial expressions, tone of voice, shared environments, and social accountability all provide signals that help people evaluate trustworthiness.
Online environments, however, change many of these dynamics.
Digital interactions often lack physical cues, are mediated by technological systems, and can occur between strangers who have never met. As a result, users rely more heavily on symbolic indicators of credibility, such as visual design, institutional branding, perceived authority, or recommendations.
This shift already made trust more fragile online. But the emergence of generative AI is changing the landscape even further.
Generative AI and the Manipulation of Goals
From the perspective of the socio-cognitive model, generative AI introduces new possibilities for influencing the very ingredients that create trust.
The first of these ingredients is the goal.
AI technologies make it possible to identify potential targets more precisely and more quickly. In terms of trust, this means finding people who already have goals that can be exploited, particularly goals with a high perceived value, where the potential loss or failure carries significant consequences.
For example:
- Investment scams can target individuals who have already shown interest in cryptocurrencies or financial opportunities.
- Romance scams can identify people who appear to be searching for relationships.
The increasing availability of personal data and patterns of online self-disclosure also makes it easier to identify these targets — a topic discussed in a previous CyberPsy article on how digital self-disclosure can unintentionally reveal personal vulnerabilities.
However, AI can also do something more subtle: it can make a goal emerge where none previously existed.
If no goal is present, one of the fundamental ingredients of trust is missing. But attackers can introduce or amplify a goal by presenting a situation as urgent, important, or personally relevant.
This mechanism is common in phishing attacks or scams involving requests from colleagues, executives, friends, or family members. In these cases, the attacker constructs a situation — a problem to solve, an urgent payment, or a time-sensitive request — that suddenly becomes a goal for the victim.
AI technologies contribute to this process not only by identifying suitable targets through data analysis, but also by helping construct convincing scenarios. Tools for impersonation, voice cloning, and deepfake audio or video can make these situations appear real and credible.
Once a goal has been introduced or activated, the second ingredient of trust becomes essential: beliefs.
Beliefs and the Illusion of Reliable Sources
After a goal has been established, the decision to delegate a task — and therefore to trust — depends on beliefs about the other agent.
Generative AI can influence these beliefs in multiple ways.
One key factor is the source of the beliefs. When a message appears to come from a credible or familiar source, the perceived validity and certainty of the information increases.
AI can now convincingly simulate such sources through impersonation, voice synthesis, or realistic video and audio deepfakes. Even simpler techniques, such as perfectly crafted emails or cloned websites, can make a source appear legitimate.
But the influence of AI does not stop at appearance.
Research has already shown4 that large language models can outperform humans in certain online persuasive contexts, particularly when messages are personalized through micro-targeting. As discussed in a previous CyberPsy article on persuasion in the age of generative AI, conversational systems can adapt their tone, arguments, and style to individual users. When responses are coherent, personalized, and context-aware, people are more likely to perceive the source as credible — reinforcing the beliefs that form the basis of trust.
Through personalization and conversational interaction, generative AI can strengthen several key beliefs involved in the construction of trust:
- Belief in competence: fluent and context-appropriate language suggests knowledge and capability.
- Belief in willingness: rapid, cooperative responses create the impression of helpfulness and motivation.
- Belief in unharmfulness: polite and measured communication reduces perceived threat.
- Belief in context: convincing emails, SMS messages, or websites create the impression of a legitimate environment.
- Belief in dependence: the victim may come to believe that only the interlocutor can solve the problem.
Together, these elements can lead individuals to treat the AI-mediated source as trustworthy.
Rethinking Trust in the AI Era
The socio-cognitive model of trust reminds us that trust does not emerge from a single factor. It results from the interaction of goals, beliefs, risks, and sources of information.
Generative AI does not fundamentally change the psychology of trust. What it changes is the scale, speed, and precision with which these ingredients can be manipulated.
Attackers can now identify people with valuable goals, create goals where none existed, and reinforce beliefs using highly personalized and convincing communication.
Understanding how trust is constructed — and how its ingredients can be influenced — is therefore becoming increasingly important in the digital age.
Because in a world where artificial agents can convincingly imitate humans, the challenge is no longer simply whether we trust. It is how that trust is being built — and by whom.
Further reading
If you’d like to explore some of the ideas discussed in this article:
Trust Theory
Castelfranchi and Falcone (2010) present a socio-cognitive model explaining how goals, beliefs, risk, and sources interact in the formation of trust.
Trust Theory: A Socio-Cognitive and Computational Model, 2010.
Trust in Social Systems
Niklas Luhmann explores how trust functions as a mechanism that allows individuals to cope with complexity in modern societies.
Trust and Power, 1979.
AI Models Surpass Humans in Online Persuasive Debates
A study published in Nature Human Behaviour demonstrating that advanced AI models can outperform humans in persuasive online debates when personalization is applied. https://www.nature.com/articles/s41562-025-02194-6