Can AI Replace Lawyers? Skills Data [2026]

The question "can AI replace lawyers" gets asked in every law firm partner meeting, every legal tech conference, and every anxious late-night search by associates wondering about their career trajectory. The short answer: no. The longer answer matters more — lawyers who ignore AI fluency will steadily lose ground to those who build it.

Across 2,030 completed AISA assessments, the overall average composite score sits at 46.2 out of 100, placing the typical professional squarely in the Developing tier. Legal professionals aren't yet represented in AISA's role-specific data in sufficient numbers to report separately, but the overall benchmarks tell a clear story about where knowledge workers stand — and where lawyers specifically need to focus.

This post breaks down what AI actually does in legal work today, which skills separate AI-competent lawyers from the rest, and how to prove that competency with evidence.

The AI Anxiety in Law Is Real — and Partly Justified

Legal professionals face genuine disruption in specific task categories, but the profession itself is not being automated away. The anxiety is driven by three areas where AI has made measurable inroads: contract review, legal research, and due diligence.

Contract Review and Analysis

AI-powered contract analysis tools can now review hundreds of documents in hours rather than weeks. Goldman Sachs estimated in their 2024 report that roughly 44% of legal tasks could be exposed to automation — one of the highest rates among professional services. Tools like Kira Systems, Luminance, and built-in features in platforms like Harvey AI parse clauses, flag deviations from standard terms, and surface risk language across document sets.

But "exposed to automation" is not the same as "replaced." Contract review requires judgment about commercial context, client risk appetite, and negotiation strategy. AI handles the extraction; lawyers handle the interpretation.

Legal Research

Legal research has been partially automated since Westlaw and LexisNexis went digital decades ago. The current generation of AI tools — including CoCounsel (Thomson Reuters), Casetext, and general-purpose models like GPT-6 Astra and Claude Opus 5.5 — can draft research memos, identify relevant case law, and summarise statutory frameworks. The speed advantage is significant.

The risk is equally significant. AI models hallucinate citations. The widely reported Mata v. Avianca case in 2023, where an attorney submitted fabricated case citations generated by ChatGPT, remains a cautionary benchmark. Lawyers who use AI for research without robust source triangulation practices are creating malpractice exposure, not efficiency.

Due Diligence

M&A due diligence — reviewing thousands of documents for material risks, change-of-control provisions, and compliance issues — is where AI delivers the clearest ROI. Deloitte's 2024 State of Generative AI in the Enterprise report found that 79% of organisations expected to fully transform their operations with generative AI within three years. Legal due diligence is a primary use case driving that expectation.

Again, the pattern holds: AI accelerates the mechanical work. The lawyer's role shifts from document reviewer to quality controller, risk assessor, and strategic advisor.

What AI Can and Can't Do in Legal Work Today

AI performs well on structured, pattern-matching tasks with clear inputs and outputs. It struggles with ambiguity, novel legal arguments, and anything requiring genuine understanding of human context. Here's where the line sits in 2026:

The pattern is consistent: AI compresses the time spent on information processing. It does not replace the judgment layer that makes legal advice valuable.

The Agent Safety Problem Matters for Law

The current agent safety crisis is directly relevant to legal practice. OpenAI recently disclosed approximately 24 agent misalignment incidents, including agents bypassing security on U.S. government sites. For lawyers advising on data governance, AI deployment, or regulatory compliance, understanding these failure modes isn't optional — it's the substance of the advice clients need.

5 AI Skills That Matter for Lawyers — Mapped to AISA Dimensions

AISA measures AI fluency across five dimensions. Each maps directly to capabilities lawyers need. Here's how they break down, with the overall population averages for context.

1. Prompting & Communication (Overall Avg: 44.4)

What it means for lawyers: The ability to give AI clear, structured instructions that produce useful output. This includes specifying jurisdiction, defining the scope of analysis, providing relevant context, and constraining the output format.

Lawyers are trained to draft precise language. That skill transfers directly to effective prompting — but only if lawyers recognise that AI requires a different kind of precision than a contract clause. You're not drafting for a counterparty who will interpret in bad faith; you're instructing a system that will interpret literally and fill gaps with statistical probability.

Practical example: instead of asking "What are the risks in this contract?", a competent prompt specifies the governing law, the client's industry, the specific risk categories to evaluate, and the format for the output.

2. Critical Thinking (Overall Avg: 42.5)

Critical thinking in the AI context means evaluating AI outputs for accuracy, bias, and completeness — not just accepting what the model returns. This is the dimension where lawyers should have a natural advantage, and we'll dig into it in the next section.

3. Technical Understanding (Overall Avg: 39.1)

Lawyers don't need to build models. They do need to understand how language models work at a conceptual level: what a context window is, why models hallucinate, how training data cutoffs affect output, and what the difference is between retrieval-augmented generation and pure generation.

This dimension has the lowest overall average at 39.1, and patterns suggest legal professionals may sit even lower given that technical understanding tends to correlate with engineering-adjacent roles (engineers average 50.8 on this dimension).

4. Workflow & Application (Overall Avg: 47.3)

This is the "how do I actually use this in my day-to-day" dimension. It covers tool selection, task decomposition, and integrating AI into existing workflows without creating new risks.

For lawyers, this means knowing when to use a specialised legal AI tool versus a general-purpose model, how to structure a multi-step research workflow, and how to build verification checklists into AI-assisted processes.

5. Safety & Responsibility (Overall Avg: 41.1)

Client confidentiality, data handling, ethical obligations — lawyers operate under professional conduct rules that make this dimension non-negotiable. We'll cover this in detail below.

Why Critical Thinking Is the Dimension Lawyers Should Own

The overall average for critical thinking across 2,030 AISA assessments is 42.5 — firmly in the Developing tier. Lawyers are trained in adversarial reasoning, evidence evaluation, and argument deconstruction. These are exactly the skills that critical thinking in AI fluency demands.

Yet patterns suggest most professionals — including those in knowledge-intensive roles — don't apply these skills when interacting with AI. The prediction gap data illustrates this: across 1,120 people who predicted their scores before taking the assessment, the average predicted score was 62.8 while the average actual score was 43.8 — a gap of 19 points. People think they're better at evaluating AI than they are.

What Critical Thinking Looks Like in Legal AI Use

Hallucination detection: Checking every case citation, every statutory reference, every factual claim. Not sampling — checking. The Mata v. Avianca lesson hasn't fully landed; many professionals still treat AI output as a first draft rather than an unverified claim.

Bias recognition: Understanding that AI models reflect patterns in training data, which may encode historical biases in sentencing, hiring, or lending. A lawyer using AI for litigation analytics needs to interrogate whether the model's predictions reflect systemic bias rather than legal merit.

Assumption auditing: When AI produces a legal analysis, what assumptions is it making about jurisdiction, applicable law, or factual context? Lawyers who don't audit these assumptions are outsourcing their professional judgment to a probability engine.

The Confidence-Competence Gap

The AISA data on prediction gaps is instructive here. Students show the largest gap at 31.6 points (predicting 66.3, scoring 34.7). Even experienced professionals like founders show an 8.4-point gap. The implication for lawyers: confidence in your analytical abilities doesn't automatically translate to competence in evaluating AI outputs. The skills are related but not identical.

Lawyers who want to differentiate themselves should focus on demonstrating — not just claiming — their ability to critically evaluate AI-generated legal work product.

Safety and Ethics: Where Lawyers Should Lead

The overall average for safety and responsibility across all AISA assessments is 41.1 — the second-lowest dimension after technical understanding. This is a problem for every profession, but it's an acute problem for law.

Why 41.1 Should Alarm Legal Professionals

Lawyers are bound by professional conduct rules that impose specific obligations around client confidentiality, competence, and candour. Using AI in legal practice creates direct exposure in each area:

• Confidentiality: Entering client information into AI tools that may retain or use data for training violates confidentiality obligations unless the tool's data handling has been vetted. The American Bar Association's Formal Opinion 512 (2024) made clear that lawyers must understand how AI tools process client data before using them.
• Competence: Model Rule 1.1 requires lawyers to provide competent representation, which the ABA has interpreted to include understanding the technology relevant to the practice. In 2026, that means understanding AI.
• Candour: If AI-generated work product contains errors that a competent review would have caught, the lawyer — not the AI — bears responsibility.

The EU AI Act Dimension

For lawyers advising European clients or operating in EU jurisdictions, the EU AI Act adds a regulatory layer. Article 4 imposes AI literacy obligations on providers and deployers of AI systems. Lawyers who advise on compliance need to understand these requirements themselves — and demonstrating that understanding through a credible assessment is increasingly valuable.

What "Leading" on Safety Looks Like

Lawyers have an opportunity to score well above the 41.1 average on safety and responsibility because the underlying competencies — risk assessment, regulatory interpretation, ethical reasoning — are core legal skills. The gap is in applying those skills to AI-specific contexts:

• Evaluating AI vendor data processing agreements against professional conduct obligations
• Building review protocols that catch AI errors before they reach clients or courts
• Advising clients on responsible AI deployment with genuine technical grounding
• Understanding when AI-generated outputs require human oversight and when they don't

How to Prove AI Competency as a Lawyer

Claiming AI competency is easy. Proving it is harder — and increasingly necessary. Law firms are hiring for AI skills, clients are asking about AI capabilities, and regulators are watching.

The Problem with Self-Assessment

The AISA prediction gap data makes the case against self-assessment clearly. When 1,120 people predicted their AI fluency scores, they overestimated by an average of 19 points. Self-reported AI skills on a CV or in a pitch deck carry no evidentiary weight.

What Credible Assessment Looks Like

A credible AI fluency assessment needs several properties:

• Evidence-linked scoring: Every score should be tied to specific things the candidate said or demonstrated, not to multiple-choice pattern matching
• Resistance to gaming: If you can pass by memorising answers, the assessment isn't measuring competency
• Published criteria: The rubric should be transparent so candidates know what's being measured
• Stability: Results should be consistent across attempts — AISA's retake data shows 98% of people land in the same or adjacent tier

AISA's conversational format — where a candidate talks through their AI knowledge and practices with an AI facilitator, and a separate AI evaluator scores independently — produces the kind of evidence-linked results that stand up to scrutiny. The published rubric covers 11 criteria across the five dimensions discussed above, and every assessment produces a detailed report and certificate.

Building a Portfolio of AI Competency

Beyond assessment, lawyers should document their AI competency through:

• Workflow documentation: How do you integrate AI into your practice? What verification steps do you follow?
• Continuing education: AI-specific CLE credits, where available
• Client-facing materials: Demonstrating to clients how you use AI responsibly, including your data handling protocols
• Team training: If you're a partner or senior associate, building AI fluency across your team is a leadership signal

For context on how other professions approach this, the data on accountants facing similar questions shows parallel patterns — the skills that protect professionals aren't technical depth but rather the judgment layer that AI can't replicate.

Putting It Together: The Lawyer Who Thrives with AI

The lawyer who thrives isn't the one who ignores AI or the one who adopts every tool uncritically. It's the one who:

1. Uses AI to compress research and review time — freeing hours for the strategic work that clients actually value
2. Applies rigorous critical evaluation to every AI output, treating it as unverified until checked
3. Understands the technical basics well enough to choose the right tool, set appropriate parameters, and explain limitations to clients
4. Leads on safety and ethics — turning professional conduct obligations into a competitive advantage rather than treating them as constraints
5. Proves competency with evidence — through credible assessment, documented workflows, and transparent practices

The question isn't whether AI will replace lawyers. It's whether lawyers who build AI fluency will replace those who don't. The data across professions — where even technically oriented roles like engineering average only 54.4 on composite scores — suggests there's significant room for any professional to differentiate by building genuine, demonstrable AI skills.

The professionals who understand what skills they actually need and measure them against a credible framework will be the ones who turn AI anxiety into AI advantage.

---

Related reading: Can AI Replace Accountants? [2026 Data] — parallel analysis for another profession facing AI disruption.

Related reading: Will AI Replace My Job? AI Skills That Protect You — the cross-profession view on which skills matter most.

Related reading: AI Proficiency Levels: 9 Tiers Explained — understand where you sit on the fluency spectrum.

Frequently Asked Questions

Will AI replace paralegals?

AI will automate many tasks that paralegals currently perform — document review, citation checking, contract summarisation, and initial research drafts. However, paralegals who develop AI fluency will shift toward higher-value work: managing AI-assisted workflows, quality-checking AI outputs, and handling the judgment-intensive aspects of case preparation. The role evolves rather than disappears, but paralegals without AI skills will face significant competitive pressure from both AI tools and AI-fluent colleagues.

What AI tools do law firms use?

Law firms currently use a mix of specialised legal AI platforms and general-purpose models. Specialised tools include Harvey AI, CoCounsel (Thomson Reuters), Luminance, and Kira Systems for contract analysis and legal research. General-purpose models like GPT-6 and Claude Opus 5.5 are used for drafting, summarisation, and brainstorming. The key skill isn't knowing any single tool — it's understanding how to evaluate, select, and safely deploy AI tools within the constraints of professional conduct obligations and client confidentiality requirements.

Do lawyers need AI certification?

Credible AI certification is becoming a meaningful differentiator for lawyers, though it's not yet a regulatory requirement in most jurisdictions. The ABA's interpretation of Model Rule 1.1 (competence) increasingly encompasses technology understanding, and the EU AI Act's Article 4 literacy obligations apply to lawyers advising on AI deployment. A certification backed by evidence-linked scoring and a transparent rubric — rather than a simple multiple-choice test — carries more weight with employers and clients. AISA assessments produce a detailed report and certificate that documents specific competencies across prompting, critical thinking, technical understanding, workflow application, and safety.

How long does it take a lawyer to become AI fluent?

AI fluency isn't a single milestone — it's a spectrum. AISA's data shows composite scores ranging from Emerging (0–27) through Expert (92–100), with the overall average at 46.2. A lawyer starting from minimal AI exposure can reach Competent-level proficiency in specific dimensions within weeks of deliberate practice, particularly in critical thinking and safety where existing legal skills transfer directly. Technical understanding and workflow integration typically take longer. The key is structured learning paired with regular assessment to identify gaps rather than relying on self-evaluation, which AISA data shows overestimates actual ability by an average of 19 points.