本文目前仅提供英文版本。
AGI Readiness for the Legal Profession: Governance When AI Approaches Expert Legal Capability
Large language models already generate legally plausible content. As AI approaches expert legal capability in research, drafting, and analysis, the governance implications for law firms, in-house teams, and legal regulators are profound. The readiness guide.
Key Takeaways
AI is already transforming legal practice in document review, contract analysis, legal research, and drafting. The governance challenge is not whether to use AI but how to use it in ways that satisfy professional obligations and serve clients well.
The hallucination problem is particularly acute in legal AI — fabricated cases and invented statutory provisions have already resulted in court sanctions. Legal professional competence obligations require verification of all AI-generated legal content.
As AI approaches expert legal capability in research and drafting, the professional value proposition shifts from doing to supervising and judging — the lawyers who remain valuable are those who can evaluate AI output, exercise judgment, and maintain client relationships.
Legal regulators (SRA in the UK, state bars in the US, state law societies in Australia) are developing AI-specific conduct guidance. The trajectory is consistent: AI use is permissible with appropriate supervision, but supervision must be genuine.
Client confidentiality obligations require that legal AI tools be specifically approved for use with client data — most commercial AI tools do not meet the confidentiality standard without specific configuration and contractual protections.
"仅供参考。本文不构成法律、监管、财务或专业建议。如需具体指导，请咨询合格专家。"
Where legal AI is now and where it is going
In 2026, legal AI is well-established in several practice areas. Document review — identifying relevant documents in discovery or due diligence — has been AI-assisted for over a decade, with predictive coding reducing review costs by 50-80% on large matters. Contract analysis and comparison has moved from specialist tools to capabilities embedded in mainstream legal workflow platforms. Legal research AI can now identify relevant case law and statutory provisions with increasing reliability. And legal drafting AI — tools that generate first drafts of contracts, pleadings, correspondence, and memoranda — has become widely adopted despite significant accuracy concerns.
The trajectory of AI capability in legal work follows a consistent pattern: AI first approaches and then exceeds human performance in high-volume, pattern-matching tasks (document classification, citation identification, standard clause recognition), while remaining unreliable in tasks requiring genuine legal judgment (interpreting ambiguous provisions, assessing litigation risk, advising on strategy, exercising professional discretion). The governance challenge is that these two categories are often not clearly distinguished in legal practice — lawyers use AI for both and may not always appreciate which category they are in.
The hallucination liability
The legal AI hallucination problem — AI systems generating confident references to cases that do not exist, statutory provisions that were not enacted, or legal principles that are incorrect — is not a minor technical inconvenience. It has already resulted in court sanctions for lawyers who filed AI-generated briefs without adequate verification. The specific documented cases involve lawyers submitting citations to fabricated cases generated by ChatGPT, with courts imposing sanctions under Rule 11 (US) and equivalent provisions in other jurisdictions for filing pleadings without adequate verification of their content.
The professional competence obligation provides the governance framework: lawyers are required to exercise the care and skill of a competent practitioner. Using AI to generate legal content without adequate verification does not satisfy this standard. The verification required depends on the nature of the content — a standard commercial contract clause requires less verification than a novel legal argument about statutory interpretation — but the obligation applies in all cases. Law firms developing AI use policies must address the verification requirement specifically: who verifies AI output, using what methodology, with what documentation of that verification.