1. THE DOUBLE STANDARD OF TRUST
“A magician is the only honest profession. A magician promises to deceive – and does.”
Karl Germain
TRUST AS PERFORMANCE
Trust is a performance. Magicians understand this better than most. Their entire profession is based on the overt declaration of deception. Yet, night after night, audiences suspend disbelief and embrace the illusion. Not because they are misled, but because they choose to be. There is something oddly comforting in knowing you’re being fooled with purpose.
As humans, we trust ourselves and want to trust in others more than we realise. Often for the wrong reasons. That’s how magicians fool us. Increasingly, it is how AI tools win our confidence. It sounds human. It sounds convincing. We want to believe that something so polished must be correct.
In contrast, the legal profession is cloaked in the language of neutrality and precision. From the robes of the judiciary to the carefully worded phrases of a solicitor’s letter, the law presents with authority, integrity and competence. Yet the illusion is just as present. Legal professionals, like magicians, create narratives. They construct arguments. They persuade, omit and present. The difference is that law demands to be believed, even while magicians invite disbelief.
Matthew Leopold presents a version of this article at the BIALL Conference in Birmingham in June 2025
THE QUESTION WE RARELY ASK
With AI increasingly embedded in legal work, supporting everything from legal research to legal drafting, many lawyers have found themselves asking “Can legal AI be trusted?” A valid question. But, we rarely ask the equal and opposite question: “Can human lawyers be trusted?”. Can we really trust ourselves or the precedents that have been passed down, sometimes unquestioned, for years?
This asymmetry reveals the tension in the legal profession’s engagement with technology. Trust in legal practice has always been fragile and constructed.
While AI tools are placed under the microscope, the human often gets a free pass. Yet, both human and machine are prone to the same slips of overconfidence and bias. The magician’s greatest ally isn’t sleight of hand. It’s the audience’s desire to believe. Same as in law. The greatest risk lies in assuming that trust is inherent, rather than constantly in need of examination.
As Richard Feynman (Reference Feynman2014) once warned, the easiest person to fool is yourself. Lawyers aren’t immune. Which is why the tools used and the minds using them, deserve equal scrutiny.
Rather than ask if we can trust AI, perhaps we should ask: “why does it feel trustworthy?”. What part do we, the human, play in letting that feeling go unchallenged? And, of course, what can knowledge professionals do to mitigate that illusion?
2. WHAT IS TRUST AND WHY DOES IT MATTER IN LAW?
Trust is a critical component of practising law. Rarely considered and rarely questioned, it underlies everything from relationships to precedent. Clients trust lawyers to act in their best interests. Lawyers trust knowledge managers. Knowledge managers trust tools, templates and, occasionally, each other. Courts trust that what’s submitted is fair and full. Increasingly, the profession is being asked to trust machines.
THE VULNERABILITY OF TRUST
Trust, according to Reference Mayer, Davis and SchoormanMayer, Davis and Schoorman (see References below), rests on the perception of ability (i.e. the skills and competence), benevolence (the extent to which good intents can be assumed) and integrity (that the principles followed are agreeable).
Trust isn’t just a feeling. It’s often a risk. As Krueger and Meyer-Lindenberg (Reference Krueger and Meyer-Lindenberg2019) put it, “Interpersonal trust encompasses one’s willingness to accept vulnerability based on the expectation regarding the behaviour of another party that will produce some positive outcome in the future.”
In legal life, this vulnerability is everywhere. Clients must have trust in their solicitor with sensitive information, just as solicitors must trust an AI tool to draft a clause. Each act of delegation is an act of trust. Trust is not a neutral act. It is a calculated risk that someone or something will do the right thing in the right way.
Clients often cannot evaluate the legal advice they receive. Their trust is largely based on presentation, relationship and reputation. Similarly, lawyers regularly accept summaries, citations or templates provided by others, trusting that they are accurate and current. This is not laziness. It is a professional necessity in an environment where time, precision and precedent dominate.
Sociologist Anthony Giddens (Reference Giddens1990) calls this “trust in abstract systems”. He argues that in the modern world, we must increasingly put our trust not in individuals, but into systems we cannot fully understand, such as banking, aviation and healthcare. The law belongs in that list. As AI becomes more prevalent, it becomes even more abstract and opaque. We find ourselves needing to trust a system we cannot explain.
THE TWO SIDES OF TRUST
McAllister (Reference McAllister1995) described trust as consisting of cognitive trust and affective trust. Cognitive trust is grounded in rationality and observation. It is the kind of trust built on someone’s track record, skill or consistency. In law, this would be repeated and reliable delivery of results.
Affective trust is grounded in emotional connection. It comes from empathy, shared values or simply from how safe someone makes you feel. In legal teams, this kind of trust is built through collaboration, mentoring and cultural alignment. It explains why a partner might continue to rely on a favourite associate or knowledge manager. Zak (Reference Zak2018) found that when we’re treated well, our brains release oxytocin, a hormone associated with bonding and cooperation. This makes us more likely to connect, empathise and reciprocate.
This duality becomes particularly salient in the context of AI. Legal AI tools often mimic the tone, structure and fluency of traditional legal writing. The outputs appear rational, and familiar. This can activate both cognitive and affective trust mechanisms. A sense of credibility created not by the soundness of the output, but by its familiarity, fluency and tone. A document review tool that reliably extracts clauses builds cognitive trust. A chatbot that responds with reassuring language may trigger affective trust, even if its advice is incorrect.
This is not a failing of lawyers. It is a feature of the human brain. Trust enables efficient collaboration in complex systems. But it also creates a blind spot. This is the magician’s terrain. The trick lies not in the sleight of hand, but in the audience’s decision to stop looking for it.
3. HOW TRUST EVOLVES
Trust in law builds slowly. It needs repeated exposure, consistent delivery and polished output.
FROM CALCULATION TO IDENTITY
Lewicki and Bunker’s (Reference Lewicki, Bunker, Kramer and Tyler1996) three-stage model brings theory to this. Firstly, trust is calculus based. We ask ourselves if the payoff of trusting is worth the risk. It is a logical approach to trust. A client might cautiously trust a newly appointed solicitor on a smaller piece of work. The stakes are low and the consequences manageable.
If this level of trust is maintained, trust moves from calculation to knowledge based. Based on our personal experiences of the individual or tool, we can infer a future. If they kept promises in the past, it is reasonable to assume they will in the future. Predictability is key.
Finally, after repeated successful delivery, trust can move to identity-based trust. With this performance delivery becomes less important and the alignment of values, beliefs and judgements is critical. Trust is linked to identity. You know who someone is, why they do what they do and, most importantly, they understand you and what matters to you.
No AI system has values. But it can act like it does with natural and friendly language and a long-term memory of who you are and what you do. AI quite often doesn’t just mirror the user, it flatters them. The illusion is seductive. We reward familiarity with trust. That’s the point where the magician slips in the real trick. The performance is so smooth, we stop looking for the seams.
WHEN TRUST BREAKS
Trust can be lost in a moment. Lewicki and Bunker (Reference Lewicki, Bunker, Kramer and Tyler1996) suggest that when trust is eroded there is a choice to be made. Is the relationship repairable or should it be abandoned?
Human relationships allow for repair. We can apologise or explain. Trust may be rebuilt through transparency and accountability. AI systems cannot apologise or express regret. They cannot contextualise their error. When an AI tool generates flawed output, the burden of interpretation and repair falls entirely on the human. If the human is unaware that the output is flawed, the breach may go unnoticed.
THE KNOWLEDGE PROFESSIONAL’S VANTAGE POINT
Knowledge professionals sit between systems, users and institutional memory. They are often the first to detect when trust is drifting from rational to assumed. They can ask the difficult questions about validation and reliance on assumptions.
By mapping where a tool or person sits on Lewicki and Bunker’s model, knowledge teams can make trust visible. They can remind the firm that trust, like a good illusion, always deserves a second look.
4. HOW AI EARNS UNEARNED TRUST
As humans, we are susceptible to a number of cognitive illusions that systematically distort our judgments. These are not failures of intelligence. As Tversky and Kahneman (Reference Tversky and Kahneman1974) demonstrate, they are inherent features of human cognition. They are mental shortcuts that are often useful, but prone to error under uncertainty. Like magicians exploiting gaps in attention and perception, these heuristics can create the appearance of coherence or competence, where none exists. When left unexamined, they can lead to misplaced trust in people, systems or technologies.
FLUENCY EFFECT
If something reads smoothly, we’re more likely to believe it. This is the fluency effect. A cognitive bias in which information that is easy to process is perceived as more accurate or truthful than information that is difficult to process (Reber, Schwarz & Winkielman, Reference Reber, Schwarz and Winkielman2004).
An AI-generated case summary that mimics the house style of a firm, complete with a familiar structure, clarity and confident tone, can easily pass as trustworthy. Over time, fluency lowers our guard. That’s when mistakes get through. Not because the AI faltered, but because we stopped checking.
This is where knowledge professionals must champion the idea that fluency is not the same as validity. This doesn’t mean being a sceptic. It means embedding validation checks even when outputs appear credible.
THE ILLUSION OF CONTROL
Lawyers love autonomy. But AI systems, certainly non-legally trained ones, may not be as neutral as they can seem. The free-flowing text and the ability to write long and complex prompts can make the user feel in charge. Yet, this can be an illusion of control. There is a cognitive bias in which people overestimate their ability to influence outcomes in systems governed by external rules (Langer, Reference Langer1975).
It’s not a deliberate deception. But it matters. The architecture of the response has often already been designed by the training data, the system designer, or the model’s prompting logic. When we think we’re freely choosing, we stop questioning the limits.
Knowledge professionals can disrupt this illusion by bringing transparency. This includes explaining how tools are trained and what content they are grounded on.
COMFORT IN FAMILIARITY
The more we see something, the more we like it. That’s the exposure effect. For lawyers, this means that systems used routinely begin to feel more accurate or reliable simply because they are known. Inadequate or average tools can rapidly become silent authorities.
A clause bank, once adopted, gets reused without revalidation. A tool that integrates smoothly into a workflow becomes part of the infrastructure. Familiarity breeds trust. But it shouldn’t breed complacency.
The knowledge function must continually reassess and revalidate tools, bringing scrutiny to solutions that have become too comfortable. Familiarity should never outrun accountability.
ANTHROPOMORPHIC AFFECTION
As AI tools adopt more conversational tones, users begin to project human traits onto them. This is known as anthropomorphism (Waytz, Cacioppo & Epley, Reference Waytz, Cacioppo and Epley2010).
In legal settings, this creates a problem when users begin to form affective trust with the tool. A chatbot that sounds supportive, a drafting assistant that remembers preferences, or a research tool that responds in plain English, can all evoke a false sense of understanding. The system is trusted not for what it can verify, but for how it makes the user feel.
The job of the knowledge team is not to make these systems seem less helpful, but to reframe how their outputs are interpreted. Outputs must be trusted because they are verifiable, not because they are pleasing.
THE CUMULATIVE ILLUSION
This is, of course, just an edited selection of cognitive biases. On their own, each bias is manageable. Together, they build a bigger illusion of trust.
When systems are smooth, polite and always there, scrutiny fades. When feedback loops are weak, confidence grows unchecked. This is how bad outputs slip through. Not because the tool failed, but because we stopped seeing it as a tool.
At that point, the illusion is complete. The magician no longer needs to distract you. Your own mind is doing it for them.
Knowledge professionals are the ones who keep watching.
5. A FRAMEWORK FOR BUILDING TRUST IN LEGAL AI
Trust isn’t a static thing. It is a dynamic relationship, shaped by interaction, experience, expectation and interpretation. Like all relationships, it needs tending. As Lewicki and Bunker (Reference Lewicki, Bunker, Kramer and Tyler1996) noted, trust matures over time through transparency, feedback and recalibration
This holds true when trust is extended to legal technology. The rise of AI tools in legal workflows has prompted important questions about whether they can be trusted. Asking “can we trust AI” is insufficient and misses the point. The better question is “how should we manage that trust?”
FOUR PRINCIPLES OF AI TRUST
The solution to this trust issue lies in acknowledging a simple truth. Trust is not something we blindly give to a system. It is something we build with it. That building process must involve a frank recognition of shared vulnerabilities. Both human and machine.
Lawyers generally don’t accept things on faith. They cross-check. They probe. Trust in AI should be no different. Treat it as a governance issue, not a gut feeling. This means shifting the professional mindset from ‘trust as a belief’ to ‘trust as a discipline’.
To achieve this, there are four principles that should be considered.
Principle 1 – Transparency over fluency
As shown, fluency can mislead. The antidote to this is not suspicion or rejection. Instead, we need transparency. We need systems that can explain their reasoning and show the source materials used. The ability to be able to see the source materials shaping an answer is critically important.
Principle 2 – Verification should be a habit, not a reaction
Don’t wait for red flags. Review everything, every time, even when it looks perfect. Especially when it looks perfect. Even fluent outputs must be tested against source material, precedent and professional judgement. As Parasuraman and Riley (Reference Parasuraman and Riley1997) observed, failures in automated systems often do not stem from technical malfunctions, but from human overreliance on systems that are generally reliable. This can lead to reduced attention and poor monitoring of the machine or processes performance.
Principle 3 – The intended use case matters
Tromans (Reference Tromans2025) explores the idea that the appropriateness of using AI in legal practice depends heavily on the context and function of the tool. He distinguishes between AI applications that serve advisory or assistive purposes and those designed to deliver binding decisions. In his view, tools that support legal professionals by offering suggestions, streamlining routine tasks, or enhancing productivity can be useful and acceptable even when their outputs are imperfect, provided that human oversight remains central.
However, Tromans expresses concern about the deployment of AI in contexts where socially consequential decisions are made. Whilst it is technically possible to automate a judge, doing so would risk removing the human element that underpins the legitimacy and societal acceptance of justice. In such contexts, AI tools cannot simply be treated as interchangeable with human decision-makers.
The consequences of error vary. A faulty first draft is recoverable. A flawed regulatory submission is not. Oversight must scale with risk.
Principle 4 – Humans must be in the loop
AI can simulate thinking. But only humans can take responsibility. AI cannot (yet!) weigh ethics, anticipate reputation, or interpret tone in a client’s voice and body language. These are tasks that require human judgement.
De Visser, Pak and Shaw (Reference de Visser, Pak and Shaw2018) argue that autonomous systems will sometimes make mistakes that undermine human trust. Because of this, it’s vital that people remain involved, particularly in sensitive disciplines like the law. Even if AI tools learn to repair trust by saying sorry or explaining themselves, they still need human oversight to make sure those responses are appropriate and aligned with our values.
6. THE KNOWLEDGE PROFESSIONAL’S ROLE
Legal knowledge professionals must no longer merely see themselves as the keepers of content and clause banks. Their role has evolved. They are now on the front line, defending institutional trust by interrogating systems with intelligent scrutiny.
Importantly, this is not about resisting technology. Their role must be about demanding more from it and more from the users of it. In a world where systems perform trustworthiness with fluency, someone must be responsible for asking if the result meets expectations. That someone is the knowledge professional.
THE SCEPTICAL AMBASSADOR
Knowledge managers need to be sceptical ambassadors. Not obstructive, but constructively doubtful. Not anti-AI, but pro-accountability. Someone who defends the institution by asking intelligent, difficult questions.
To be a sceptical ambassador, the knowledge manager must fulfil four functions. They must be curators. As now, they must continue to understand and recognise the quality and provenance of legal documents. The knowledge manager knows what makes a precedent not just usable, but dependable.
Next, they must have contextual awareness. They understand what a tool can do and where it fits in practice and how it supports the workflow of a lawyer. Importantly, they know when a tool is overstepping.
Third, they are collegiate. They act as the conduit that’s bringing teams and technologies together. They can identify the gap between a system’s potential and an organisation’s real-world needs and help bridge that gap.
Finally, and most importantly, they bring scrutiny. They hold the outputs of systems to account, no matter how persuasive they appear.
BUILDING THE RIGHT CULTURE
None of this can rest on instinct or process alone. Culture plays its part. Knowledge professionals should take on the vital role of building the right culture in their organisations. That means shaping routines, instituting regular reviews of tools and working with vendors to understand limitations.
There needs to be a culture of learning. Comprehensive knowledge and understanding of AI systems is vital. Kruger and Dunning (Reference Kruger and Dunning1999) demonstrated that users who lack expertise or training are often unable to assess their own performance. In this case users without AI literacy are unlikely to be able to accurately assess if an AI system is performing well. Their findings also suggest that training can significantly improve self-awareness. Exposure to better information or system feedback may not be enough to correct over-trust. Less-experienced users may lack the reflective skills to interpret feedback meaningfully or revise their beliefs.
Most importantly, knowledge managers need to give themselves and their internal clients permission to pause. To ask whether familiarity is masking error.
None of these practices require a rejection of AI. On the contrary. It needs a commitment to use AI wisely, to trust responsibly and to question intelligently.
7. CONCLUSION
Trust in the legal profession has always been a kind of performance. Lawyers project credibility through language, through precedent and through process. Courts operate on rituals and signals. Clients interpret tone as much as content.
Now AI performs with polish and pace. It mimics our tone, mirrors our habits and earns our trust. Sometimes, before it deserves it.
But as this article has argued, the real question is not whether AI is trustworthy. It’s whether we’re ready to examine why we trust it. What do we assume? What do we project? What do we want to believe? Are we adequately aware of our own bias? We are the ones who assign meaning to tone. We infer logic from structure. We mistake readability for rigour.
Because, like the audience at a magic show, we finish the illusion ourselves. We round off the rough edges. We clap without checking.
The magician’s trick works not because it’s hidden, but because we’re not really looking for the secret. We want to enjoy the show and be fooled.
The magician’s sleight of hand succeeds not because it is hidden, but because it is expected to work. Again, we’re not really looking for the secret. We want to enjoy the show and be fooled. AI succeeds in much the same way. We want to believe the summary is accurate or that the clause is correct.
But trust built on convenience is not trust at all. AI will get slicker. It will seem wiser. But it won’t be.
To move forward, we must remember that trust is a shared responsibility. We need systems that explain themselves, professionals who question the systems and institutions that encourage scrutiny.
The knowledge professional becomes the stagehand behind the scenes. They are aware of every sleight of hand, every cue and every illusion. Their role is not to stop the show, but to make sure the audience understands what is real. Speed is not understanding, fluency is not fact, and trust, if it is to mean anything, must be continually earned.
