Back to Articles
Artificial intelligence has already entered the legal field. This is no longer a speculative question. Around the world, lawyers, judges, court staff, legal researchers, and law firms are already using AI tools to summarise text, support drafting, organise material, assist research, and accelerate routine work. Yet the pattern emerging globally is not one of simple technological progress. It is one of uneven institutional readiness.
AI uptake is advancing faster than legal professions, courts, and professional bodies are building the governance systems required to supervise its use responsibly. UNESCO's global survey of judicial professionals captures this plainly: 44% of respondents reported already using AI tools in their work, yet only 9% had received training or institutional guidance, and a strong majority supported mandatory regulations and formal guidance frameworks. The American Bar Association has itself recognised that AI has moved from experimentation into the working infrastructure of legal practice.
Sources: UNESCO, Artificial Intelligence and the Rule of Law, Global Survey of Judicial Professionals. ABA Task Force on Law and Artificial Intelligence, Year 2 Report on the Impact of AI on the Practice of Law (December 2025).
The legal field has therefore crossed a threshold. The issue is no longer whether AI exists or whether legal actors are curious about it. The issue is whether the profession can govern the reality of AI use without weakening judgment, confidentiality, accountability, or public trust.
Part One
The Governance Shift
The global picture is already becoming harder to ignore
Major legal institutions have begun to respond. In the United States, the American Bar Association's Formal Opinion 512 makes clear that lawyers using generative AI still owe the same duties of competence, confidentiality, communication, and accuracy, and must understand the capabilities and limitations of the tools they use. In England and Wales, the Law Society has issued dedicated guidance on generative AI for solicitors, while the Bar Council has updated guidance for barristers, stressing risks such as hallucinations, bias, confidentiality, and information disorder.
Recent events have sharpened the stakes. In 2025, the UK High Court warned lawyers over fake case citations and inaccurate AI-assisted material submitted in proceedings, linking professional misuse of AI directly to risks to justice and public trust. The warning was not about whether AI exists. It was about the professional failure to verify and supervise its use. By mid-2025, over 600 cases in the United States had already involved lawyers citing non-existent authority generated by AI tools.
Africa is not outside this trend. UNESCO's consultations in Eastern Africa found high familiarity with AI among judicial professionals and significant day-to-day use, especially for summarising, drafting, and legal research. In South Africa, professional and regulatory concern has grown more visible, with public reporting indicating moves toward guidance development and stronger warnings around fake citations and professional misconduct.
This governance question is not theoretical in Zimbabwe. AI-assisted legal research tools tailored to Zimbabwean law are already entering the market. Case Rover describes itself as an AI-powered legal research tool for lawyers in Zimbabwe, using statutes, regulations, and court decisions and allowing users to verify against original documents. Zalari similarly markets AI-cited answers grounded in Zimbabwean judgments, with links back to source material for verification. In the wider region, tools such as Lexis+ AI in South Africa are also integrating AI-enabled legal drafting and research into established legal content platforms. These developments reinforce the central point of this article: the issue is no longer whether AI will reach legal practice, but whether the profession has the standards, boundaries, and oversight needed to govern its use responsibly.
From adoption to governability
The legal profession should not approach AI as a passing technology trend or as a niche concern for a handful of innovative firms. Nor should it assume that because AI tools are already in use, the profession has somehow already adjusted to them.
Use is not the same as readiness. Familiarity is not the same as governance.
The fact that a practitioner can access an AI system in seconds does not mean that the profession has settled the questions that actually matter: when such tools may be used, on what terms, with what safeguards, with what supervision, and with what continuing responsibility. The biggest mistake the profession could make at this stage would be to ask only whether AI makes lawyers faster. Speed is not the right test. The better test is governability.
Can legal institutions govern the use already taking place? Can firms, chambers, and legal-service organisations host AI use without compromising confidentiality? Can practitioners distinguish between AI assistance and the improper substitution of judgment? These are not anti-technology questions. They are the conditions under which technology can be integrated without corroding professional responsibility.
The governance gap that must be closed
The legal profession now faces what AGCIH terms the Governability Gap: the distance between the fact that AI is already in use and the institution's actual capacity to supervise and control that use. If use is already occurring, but there is not yet a profession-wide framework on confidentiality, verification, supervision, court-facing use, and acceptable limits, then the profession is operating inside a governability gap.
The problem is not that practitioners are experimenting. The problem is that experimentation may be taking place without a clear institutional container. Legal practice is bound up with fiduciary duties, ethical obligations, client trust, duty to the court, procedural fairness, and public confidence in justice institutions. An AI error in legal work does not remain a private inconvenience for long. It can become a defective pleading, a misleading citation, a compromised client communication, a confidentiality breach, or a failure in representation.
Why the profession's mandate already points here
This is not an artificial extension of the profession's existing role. Professional bodies already operate within mandates that include legal training, continuing professional standards, the regulation of practitioners, discipline, and the promotion of public trust and confidence in the justice system. AI governance sits directly within that mandate, not beside it.
Across jurisdictions, this alignment is becoming more visible. Professional bodies are issuing guidance, courts are issuing warnings, legal educators are revising curricula, and some jurisdictions are beginning to examine whether existing professional rules are adequate to the current environment. These are not peripheral technology questions. They go to the core of what it now means to practise law responsibly.
The danger of abdication masked as assistance
There is one dimension of the governance challenge that the profession must confront with particular seriousness. The greatest danger is not only that AI can be wrong. Lawyers have always worked with fallible tools, fallible sources, and fallible human judgment. The bigger risk is that professionals may begin to surrender too much of their own responsibility while retaining the outward appearance of authorship and control.
That is the true professional danger: abdication masked as assistance.
If AI-generated material is adopted too easily because it is fluent, fast, and persuasive, practitioners may begin mistaking convenience for reliability. Once that pattern becomes normal, professional judgment weakens not in a dramatic moment, but gradually, through habit. This is exactly why current international guidance continues to insist that AI should support legal work, not replace legal judgment, and why judicial concern has focused so sharply on verification and responsibility.
The profession must therefore begin from a disciplined premise: AI may assist legal work, but it does not relieve the lawyer of professional responsibility, and it does not reduce the profession's duty to preserve public trust.
Part Two
Professional Hosting Capacity
From use to hosting: a necessary distinction
If the legal profession is to move from casual AI use to responsible AI governance, it must begin by building what AGCIH's Administrative Hosting Capacity doctrine identifies as a foundational requirement: the institutional conditions needed to absorb new systems without losing administrative control, ethical oversight, or attributable responsibility.
Applied to legal practice, this becomes professional hosting capacity: the ability of legal institutions, law firms, legal-service organisations, chambers, and professional bodies to absorb AI into legal practice without losing human oversight, ethical control, confidentiality, accountability, and attributable professional judgment. In simpler terms, it is the difference between using AI and remaining in command of AI-assisted legal work.
Hosting is not a restriction. It is professional control. The legal profession should not aspire merely to participate in technological change. It should aspire to remain professionally sovereign within technological change. That means preserving control over confidentiality, review, judgment, supervision, client trust, and court integrity even as new tools enter the workflow.
The seven elements of professional hosting capacity
One of the easiest mistakes in AI-assisted practice is also one of the most dangerous: the unreflective input of sensitive or client-related material into external systems without a proper understanding of how that information is processed, stored, retained, or exposed.
AI introduces new pathways of exposure through prompts, uploaded documents, system memory, vendor terms, cloud architecture, and staff use habits. Confidentiality governance requires actual internal structures, not merely a vague instruction to be careful. Without such structures, confidentiality becomes dependent on individual caution alone. That is not governance. That is fragility.
One of the most dangerous qualities of AI output is not simply that it can be wrong, but that it can be wrong in a fluent, plausible, and professionally persuasive way. The profession must therefore build habits and standards of verification that are stronger, not weaker, when AI enters legal work.
Verification discipline means that no practitioner treats AI-generated material as professionally reliable merely because it is well written, confidently expressed, or delivered quickly. Legal authorities must still be checked, factual claims confirmed, legal reasoning interrogated, and final responsibility must still sit with the practitioner.
Legal practice depends on the existence of a responsible professional who can stand behind the work produced in their name. AI can blur this line if the profession is not careful.
Drawing on AGCIH's Attributable Professional Judgment concept, the principle is clear: AI may support the workflow, but it may not displace the lawyer's own review, reasoning, and responsibility. The professional must remain identifiable as the source of the judgment, not merely the signatory of its output.
AI in legal practice cannot be governed solely at the level of individual personal caution. It must also be governed institutionally through supervision and internal review.
This is especially important where junior lawyers, candidate legal practitioners, paralegals, and support staff may use AI tools in ways that materially affect legal work. Without clear supervisory expectations, the profession risks creating a hidden layer of unsupervised machine-assisted drafting and reasoning beneath outwardly conventional legal work.
The profession must think carefully about how AI affects the lawyer-client relationship: transparency, consent, cost, quality, communication, and trust. If AI materially influences drafting, research, document production, or service delivery, the standards of quality and professional responsibility that clients are entitled to expect must remain unchanged.
Client-facing boundaries also matter in relation to cost and fairness. If AI significantly reduces the time spent on a task, how is billing treated? If errors arise, how are responsibilities explained? These are not merely technical questions. They go to professional integrity.
Court-facing work carries heightened obligations because errors, inaccuracies, or fictitious material do not only affect a lawyer's internal workflow. They affect the administration of justice itself.
No AI-assisted court-facing work should enter proceedings unless a lawyer has independently verified its factual, legal, and procedural soundness. That principle is entirely consistent with the direction of recent judicial and professional guidance internationally, and it must be treated as non-negotiable.
Professional hosting capacity cannot be built through one conference, one circular, or one training session. AGCIH's Continuous Administration doctrine holds that governance of dynamic systems must be ongoing, adaptive, and institutionally embedded, not episodic.
The profession will need continuing education, policy refinement, review of emerging failures, updated practice protocols, and institutional mechanisms that can learn as tools evolve. A profession that treats AI governance as complete after an initial awareness phase will quickly fall behind reality.
Part Three
What a Profession-Wide Response Requires
Moving from awareness to readiness
If AI is already in legal practice, and if the real challenge is governance rather than adoption, then legal institutions cannot remain at the level of general concern. They must begin to build a structured response, one grounded not in regulatory panic, but in the profession's actual mandate: standards, competence, discipline, training, public trust, and the integrity of legal work.
A baseline understanding of current use and current risk
The first task is to understand what is already happening. This may take the form of member surveys, practitioner consultations, focused discussions with firms and sole practitioners, identification of common use cases, and mapping of current risks and areas of uncertainty. A profession cannot govern what it has not first honestly assessed.
Interim professional principles on AI in legal practice
At an early stage, the profession may not yet be ready for a detailed code or binding rules. But it can and should state its normative position, affirming that lawyers remain fully responsible for AI-assisted work, that AI does not displace professional judgment, that confidentiality obligations remain fully applicable, and that all authorities and material facts must be independently verified. These principles establish the profession's normative centre.
Continuing professional development aligned to actual duties
Training must address what AI can and cannot do, how hallucinations and false authority appear in practice, where confidentiality exposure arises, what verification actually requires, and how supervision must work when AI enters the workflow. Generic technology awareness is not enough. The training must be built around the duties practitioners actually carry, because those duties do not change simply because a new tool has entered the workflow.
Model internal AI use policies
Many practitioners and organisations know they should have an internal policy on AI, but are uncertain what it should actually contain. A model policy should answer the questions practitioners are already asking: what may be used, on what terms, for which categories of work, with what verification required, and what happens when something goes wrong. Those are governance questions, not technology questions. The profession can add real value by providing model language that firms and organisations can adapt to their own contexts.
Heightened protocols for consequential legal work
The profession should identify categories of legal work that require enhanced caution and explicit safeguards, including authorities and citations, pleadings and affidavits, legal opinions, submissions, review of evidence, and advice in urgent or high-stakes matters. Where legal work has heightened consequences, professional review standards must also be heightened.
Ongoing institutional review
A profession-wide response must not end with a circular, a toolkit, or a workshop. If the institution is serious, it must create some mechanism for continuing attention to this issue. AI in legal practice will not remain fixed. The tools will change, the patterns of misuse will change, and the expectations of clients and courts will change. This is why professional AI governance must be continuous, not episodic.
Why this matters beyond large firms
A profession-wide response should not focus only on large or commercial firms. AI in legal practice affects multiple parts of the legal ecosystem: private practitioners, sole practitioners, legal aid actors, NGO legal-service providers, training institutions, candidate legal practitioners, and ultimately the clients and communities who rely on legal services. Weakly governed AI does not distribute risk evenly. In many cases, the costs of poorly supervised legal automation are likely to fall hardest on those with the least power to detect, challenge, or absorb error. That is why the governance question is not only about technology competence. It is also about the ethical shape of legal service delivery.
The role of a governance-first institution
As legal professionals face this challenge, the question of institutional support becomes important. Not every actor brings the same kind of value. Technology vendors can explain tools. Consultants can demonstrate products. Training providers can run isolated workshops. But none of those things answers the more serious question now emerging before the legal field: how should the profession govern AI already entering practice in a way that preserves judgment, confidentiality, accountability, and public trust over time? That is not primarily a vendor question or a product question. It is a governance question.
A governance-first institution can translate global developments into local professional realities. Comparative material developed in London, Washington, or Geneva cannot simply be copied into different professional contexts without adaptation. The interpretive work in between is real work, and it requires governance expertise rather than generic technology commentary. Most importantly, a governance-first institution can help professions build readiness before reaction, addressing the governance issue before standards are undermined by habit, before court-facing failures multiply, and before practitioners assume that convenience is an adequate substitute for judgment.
Conclusion
The Profession Must Remain in Command
Artificial intelligence is entering legal practice in ways that are too practical, too ordinary, and too immediate to be treated as a distant policy topic. The question is no longer whether AI is present. The question is whether the legal profession will remain in command as AI becomes normal.
The profession must not confuse use with readiness. It must not confuse speed with competence. It must not confuse fluency with reliability. And it must not confuse assistance with the transfer of professional responsibility. The real danger is not only that AI can make mistakes. The greater danger is that legal professionals and legal institutions may begin, gradually and almost imperceptibly, to surrender too much of their own judgment while retaining the formal appearance of authorship and control. That is how professional authority weakens: not through dramatic failure, but through the normalisation of convenience without supervision, confidence without verification, and adoption without governance.
Governance is what ensures that lawyers remain responsible for what is done in their names. Governance is what keeps confidentiality from becoming an afterthought. Governance is what prevents legal work from becoming procedurally thinner while appearing technologically more advanced. In this sense, governance is not external to legal practice. It is part of what keeps legal practice recognisably professional.
The legal profession does not need to begin with the question: how can we use AI more? It should begin with a more disciplined and more important question: how do we ensure that as AI enters legal practice, the profession remains ethically grounded, administratively capable, and unmistakably in command of its own judgment?
That is the question now before the profession. It is also the right place to begin.
AGCIH Analytical Concepts
The following concepts are drawn from AGCIH's governance doctrine and applied in this article to the legal profession context. All concepts carry explicit AGCIH attribution.
| Administrative Hosting Capacity | (AGCIH doctrine) The institutional conditions required for a governance body to absorb an AI system into its operational and administrative functions without losing supervisory control, ethical accountability, or the capacity for attributable decision-making. Applied in this article to legal institutions as Professional Hosting Capacity. |
| Professional Hosting Capacity | (AGCIH, legal sector application) The ability of legal institutions, law firms, legal-service organisations, chambers, and professional bodies to absorb AI into legal practice without losing human oversight, ethical control, confidentiality, accountability, and attributable professional judgment. |
| Continuous Administration | (AGCIH doctrine) The ongoing institutional responsibility to supervise, update, review, and govern AI use over time, rather than treating governance as a one-off intervention. |
| Attributable Professional Judgment | (AGCIH concept) Judgment that remains clearly traceable to a responsible legal practitioner, even where AI tools assist in the workflow. The principle that authorship of legal work must carry genuine, reviewable human judgment behind it. |
| Governability Gap | (AGCIH concept) The gap between the fact that AI is already being used within an institution or profession and the institution's actual capacity to supervise and control that use responsibly. |
| Episodic Governance | A weak governance pattern in which institutions respond to AI through isolated workshops, statements, or events without creating ongoing oversight structures. The absence of Continuous Administration. |
| Assistance / Substitution Boundary | The line between AI as a support tool and AI as an improper replacement for human legal reasoning, verification, or professional responsibility. |
| Confidentiality Exposure by Design | Risk created when practitioners input sensitive information into tools or workflows that have not been institutionally assessed for confidentiality and data-governance safeguards. |
| Governance-First Adoption | An approach that does not begin by asking how to use AI more, but by asking what safeguards, standards, and oversight are required around use already taking place. |