What GPT-4 Did Not Change About Enterprise Legal Tech
Two years after the generative AI inflection point reshaped expectations about legal technology, the fundamental constraint in enterprise legal technology is unchanged: trust, accountability, and defensible outputs. The technology has improved dramatically. The procurement process has not.
To be clear about what has changed: the capability surface for legal AI has expanded enormously since early 2023. Models can now synthesize complex contracts, identify non-standard provisions across multiple governing frameworks, draft initial position papers on novel legal questions, and explain their reasoning in formats that attorneys find useful. The accuracy on well-defined tasks — specific contract clause extraction, standard document classification, routine legal research — has improved to the point where AI is genuinely faster and comparably accurate to a junior associate on those tasks.
What has not changed: a law firm partner's professional responsibility to her client does not diminish because she used AI to assist in the work. When a client loses a case because relevant precedent was missed in research, or a contract fails because a material provision was overlooked in review, the fact that AI was used in the workflow is not a defense. This accountability structure has not changed with GPT-4, and it will not change with GPT-5. It is not a technology constraint — it is a professional structure that exists to protect clients.
The implication for legal AI companies: the products that are winning in 2025 are not the ones with the best models. They are the ones that have built the accountability infrastructure around their models — citation verification that connects AI analysis to retrievable legal authority, supervision interfaces that make attorney review efficient rather than perfunctory, audit trails that document what the AI had access to and what it concluded. These are not features. They are the product. Founders who understand this are building companies that can survive a malpractice scrutiny process. Founders who do not are building demos.