← Adoption Challenges 🕐 9 min read
Adoption Challenges

AI Literacy Frameworks by Industry: What Lawyers, Physicians, CPAs, and Manufacturers Are Required to Know

Brandon Sneider · April 2026

Despite different audiences, the four frameworks converge on the same three competency themes:

See also (wiki): wiki/industry-ai-competency-frameworks.md

Executive Summary

  • Four industries have published formal AI literacy guidance through their professional bodies. Only one — legal — has crossed into binding requirements. The rest are advisory but cascading toward enforcement.
  • The ABA’s Formal Opinion 512 (July 2024) makes AI competence a malpractice question for ~1.3M US lawyers via state bar adoption. “Uncritical reliance on content created by a GAI tool is risky — and almost certainly malpractice.”
  • AAMC’s competency framework for medical educators (October 2025) and the AMA’s AI literacy policy (2024, reaffirmed 2025) are positioned to flow into LCME accreditation and CME requirements. Five pillars: human-centered, ethical, equitable, continuous, interdisciplinary.
  • AICPA’s 2025 AI in Accounting Report positions CPAs as AI assurance providers — a new service line, not just a tool-user category. 88% of finance professionals call AI the most transformative trend; 8% feel prepared.
  • Manufacturing USA’s 2025 Occupation and Competency Framework (August 2025) is the only sector building competency at the task level — micro-credentials in data annotation, manufacturing analytics, and AI-related skills via the MxD/SACA partnership.
  • None of the four frameworks specify minimum training hours. BCG’s “5+ hours” benchmark remains the only quantitative floor in the literature. Designing your program means choosing your own threshold and defending it.

The Pattern Across All Four Industries

Despite different audiences, the four frameworks converge on the same three competency themes:

  1. Understand the capabilities and limitations of the specific AI tool in use.
  2. Verify outputs through domain-specific methods — citation checking for lawyers, clinical judgment for physicians, professional skepticism for CPAs, sensor cross-validation for manufacturing.
  3. Maintain accountability to the client, patient, regulator, or end-user regardless of what the AI produced.

This is not a coincidence. Every regulated profession arrives at the same answer: AI does not transfer accountability. The professional remains liable. The framework is how each industry teaches that to its workforce.

The implication for any deployment: the curriculum content varies by role; the verification ritual — the moment a human confirms before the output goes out the door — is the universal control. Designing it well is what separates the 5% from the 95%.

Legal: The Only Binding Framework

The American Bar Association’s Formal Opinion 512 (July 29, 2024) is the only AI competency requirement with enforcement teeth. It interprets Model Rule 1.1 (Competence) and five other Model Rules to require lawyers to understand the “benefits and risks” of any generative AI tool they use to deliver legal services.

Three operational requirements stand out:

  • Lawyers “need not become GAI experts” but must have “a reasonable understanding of the capabilities and limitations” of the specific tool — and revisit that understanding as the technology changes.
  • Verification of AI outputs is “factually specific and will necessarily depend on the GAI tool and the specific task that it performs.” There is no universal verification rule; lawyers must build a per-task protocol.
  • Confidentiality (Rule 1.6), client communication (Rule 1.4), candor to the tribunal (Rules 3.1, 3.3), and supervisory duties (Rules 5.1, 5.3) all extend to AI use. Putting client data into a public model is a confidentiality breach. Filing AI-drafted briefs without verification has produced multiple sanctions in 2024–2025.

State bars in California, Florida, New York, New Jersey, and DC have followed with parallel guidance. ABA opinions are persuasive, not binding, but state bars adopting them creates the enforcement pathway. For law firms, this is no longer optional training — it is malpractice insurance posture.

Healthcare: The Frameworks Are In Place; Enforcement Lags

The AAMC’s Principles for Responsible Use of AI in Medical Education reached version 2.0 on July 31, 2025. The companion AI Competencies for Medical Educators (last updated October 2025) is mapped to the 2025 International Advisory Committee for AI Vision and Integration Framework.

The five pillars: human-centered approach, ethical use, equitable access, continuous education, interdisciplinary curriculum development.

The AMA’s AI literacy policy (adopted June 2024, reaffirmed 2025) directs the association to “develop and disseminate model AI learning objectives and curricular toolkits” for clinical practice. The AMA Ed Hub offers a free seven-part series covering ethics, evidence, and applications.

What is not yet in place: binding requirements at the practicing-physician level. CME requirements vary by state board. LCME accreditation includes AI-related expectations but does not yet specify hour minimums. Hospital systems deploying AI for clinical documentation, diagnostic support, or revenue cycle are largely setting their own training thresholds — and the deployments where workflow integration was treated as a one-time onboarding rather than a continuing competency are the ones producing the headline failure modes.

For a hospital deploying clinical AI, the AAMC framework is the credible reference point. Building a curriculum that treats it as the floor — not the ceiling — is the defensible posture.

Financial Services: Competency Becomes A Service Line

The AICPA and CPA.com 2025 AI in Accounting Report (June 2025) is unusual in two ways. First, it names specific competencies CPAs need: critical thinking, data security, AI ethics, advisory skills, and creativity alongside data analytics. Second, it positions CPAs as AI assurance providers — independent evaluators of AI systems for clients — not just tool-users.

This matters because it reframes AI literacy from a defensive posture (don’t get sued) to an offensive one (build a new revenue line). The CPA Evolution model exam (effective 2024) integrates technology and analytics as a core domain, which means the next generation of CPAs is being credentialed against AI competency at the licensing stage.

The companion data is sobering: 88% of finance professionals in the AICPA/CIMA Future-Ready Finance survey call AI the most transformative trend in accounting/finance. Only 8% feel “very well prepared.” That 80-point gap is the addressable training market.

Parallel guidance exists for investment professionals through the CFA Institute’s Code of Ethics and Standards interpretive notes, and for banks through OCC/Federal Reserve SR 11-7 model risk management guidance (covered separately in this corpus). Together they form the most mature regulatory posture of any industry covered here.

Manufacturing: The Only Sector Building At The Task Level

The 2025 Manufacturing USA Occupation and Competency Framework (released August 28, 2025) is the federal government’s attempt to standardize advanced manufacturing competencies across 17 federally-chartered manufacturing innovation institutes. It covers 18 technology areas and maps entry-level occupations to required knowledge, skills, and abilities.

What distinguishes the manufacturing approach: competencies are built at the task level, not the role level. The MxD/SACA partnership (announced December 2025) launched three co-developed micro-credentials in (1) data annotation, (2) manufacturing analytics, and (3) AI-related skills. A plant floor supervisor doesn’t need a “manufacturing AI” certification — they need to know how to label sensor data correctly, interpret an analytics dashboard, and recognize when a model output disagrees with operator judgment.

This is the most promising design pattern in the four-industry comparison. It treats AI competency as a stack of small, verifiable skills rather than a credential. For mid-market manufacturers (200–2,000 employees), the implication is direct: build training around specific tasks workers actually perform with AI, not around general “AI literacy” — which is too vague to evaluate and too broad to budget.

Adoption data on the new MxD micro-credentials is not yet available. The framework is months old.

Key Data Points

Industry Framework Issuer Date Binding? Sample/Reach Credibility
Legal Formal Opinion 512 ABA Standing Committee on Ethics Jul 29, 2024 Via state bar adoption (CA, FL, NY, NJ, DC) ~1.3M US lawyers HIGH
Healthcare AI Competencies for Medical Educators AAMC (CGEA Faculty Dev SIG) Oct 2025 (v2.0 principles Jul 2025) Advisory; cascades via LCME 155 US MD-granting medical schools HIGH
Healthcare AI in Health Care policy + 7-part Ed Hub series AMA Jun 2024, reaffirmed 2025 Advisory 270K+ AMA members MED-HIGH
Financial services 2025 AI in Accounting Report AICPA + CPA.com Jun 2025 Influencing CPA Evolution exam (2024+) ~700K US CPAs HIGH
Financial services Future-Ready Finance Survey AICPA + CIMA 2025 Survey 88% call AI most transformative; 8% feel prepared HIGH
Manufacturing 2025 Occupation and Competency Framework Manufacturing USA Aug 28, 2025 Federal-chartered standard 17 institutes, 18 tech areas HIGH
Manufacturing MxD/SACA micro-credentials (data annotation, analytics, AI) MxD + SACA Dec 2025 Voluntary Adoption data pending MEDIUM (new)
HR SHRM Body of Applied Skills and Knowledge (BASK) SHRM Updated 2024–2025 Advisory (basis for SHRM-CP/SCP) 340K+ SHRM members HIGH
HR Talent Trends 2025 SHRM 2025 Survey AI = 31% of new tech skills required in postings HIGH

What This Means for Your Organization

If you operate in a regulated industry, the framework already exists. The risk is not “what should our AI training cover” — that is answered by your professional body. The risk is whether your deployment treats the framework as the floor (defensible) or the ceiling (insufficient). Lawyers learning that the firm’s AI tool can hallucinate citations is table stakes; lawyers building per-matter verification rituals is what the ABA opinion actually requires. The same gap exists in every industry covered here.

If you operate in a sector without a binding framework — most of the economy — you have a different problem and a different opportunity. The honest answer is that you must design your own competency model and be able to defend it to a regulator, an insurer, or a board. Borrowing the structure from the closest analog is the fastest path: a professional services firm should look at AICPA’s competency list; a hospital system at AAMC’s five pillars; a manufacturer at the Manufacturing USA framework’s task-level micro-credentials. Borrowing the structure does not mean borrowing the content. The content has to fit the actual tasks your workforce performs.

The deeper pattern across all four industries: AI does not transfer accountability. The professional remains liable for the output. The verification ritual is the universal control. Designing it into the workflow — not bolting it on after a pilot ships — is what separates organizations that capture AI value from those that produce 98% more output and zero net throughput. If this raised questions specific to how your workforce should be trained against AI tools you have already deployed or are evaluating, I’d welcome the conversation — brandon@brandonsneider.com.

Sources

Brandon Sneider | brandon@brandonsneider.com April 2026