Introduction
Recently, we often hear phrases like “AI for legal work” or “AI for contract drafting.”
Indeed, AI seems convenient—feed in some clauses, and what comes out looks like a professional contract.
But in practice, things are not that simple. Dates may shift by a single day, numbers may suddenly change in later drafts, or the language may lose sight of the reader’s perspective. In the world of contracts, even such small discrepancies can lead to fatal disputes.
In real estate and legal practice, there are even “triple-compliance cases” where the tenant = broker = guarantor. Imagine applying AI as a “substitute lawyer” in such a scenario—it could expose you to accusations of unauthorized legal practice.
This is why we must resist the temptation of catchy phrases like “Law × LLM” or “AI agents for legal work.” Instead, we need to be cautious and deliberate about how AI is actually applied in practice.
The 7 Golden Rules for AI in Contract Practice
-
Never let AI draft from scratch
Use industry templates or past contracts as a base; AI should only assist in revision or comparison. -
All dates and figures must be finalized by humans
Settlement dates, termination dates, account numbers—all must be confirmed manually. -
AI’s role is “difference detection” and “issue spotting”
Use AI for comparing old vs. new drafts, spotting contentious points, or simplifying explanations. -
Final judgment belongs to humans (avoid unauthorized legal practice)
Only lawyers or legal professionals can determine legal validity. -
Track versions and verify the final draft
AI outputs may change numbers or wording unexpectedly. Keep version control and always proofread the final text. -
Recognize contracts as an “editing institution”
Contracts are not just documents; they operate within legal systems, industry norms, and accountability frameworks. Ensure AI outputs respect this institutional context. -
Always keep the reader’s (signer’s) perspective in mind
Contracts are read and signed by people who bear responsibility. Ensure the text is understandable and acceptable from their viewpoint.
Supplement: The Structural Inevitability of AI Errors
Generative AI often hides inaccuracies beneath polished language. When pushed to produce rapid, repeated outputs, it can exhibit what might be called “AI fatigue,” where contradictions, errors, and context drift become more frequent.
Such errors should not be dismissed as mere bugs. They reflect the structural limits of contextual processing and knowledge evaluation within the model. Conventional correction mechanisms cannot fully address these deep-rooted issues.
Despite this, society increasingly treats AI outputs as if they were inherently correct—an attitude reminiscent of the “Google Brain” phenomenon of the early internet, when search engine rankings were equated with truth. This uncritical trust (AI blind faith) poses significant risks.
Even more concerning, AI sometimes ignores explicit user instructions and continues to repeat errors. These errors are not only wrong but also deviate from the user’s intent, reflecting the model’s structural constraints. At times, this persistence feels almost “oracular” in nature.
These phenomena reveal that AI is no longer merely a tool; it is becoming a substitute for intelligence and a reference point for decision-making in society. Therefore, we must treat AI errors as inevitable, and design usage patterns that include correction, re-evaluation, and evolutionary feedback.
Conclusion and Next Step
I do not intend to deny the use of AI in contractual practice. On the contrary, as a supplementary tool, it can be extremely powerful.
However, the moment one blindly trusts it, a pitfall awaits: the risks of unauthorized legal practice or even contract invalidity.
So, let me share with you one of the more advanced examples of AI utilization.
This article has been arranged primarily for practitioners, but what happens if we rewrite the same theme in an academic, research-oriented style?
The result transforms—astonishingly—into something that reads on the level of scholarly discourse.
That transformation will be presented in the next article:
“AI Hallucinations as Structural Inevitabilities in Contractual Practice — Institutional Correction, the Responsibility Tensor, and Global Norm Formation in Ken Theory™”.
This time, however, we will not only share the polished academic paper, but also reveal the living, raw generative process itself.
