AI can review contracts and flag risky clauses automatically, and this is one of the more mature applications of large language models in legal workflows, though it should support rather than replace a lawyer's judgment on material agreements. A contract review system typically compares clauses against a defined playbook of acceptable and unacceptable positions, such as indemnification caps, liability limits, termination rights and auto-renewal terms, then flags deviations with a severity rating and a suggested redline or fallback position. Large language models are effective at finding unusual or missing clauses across long documents faster than manual review, and at explaining in plain language why a clause is risky, which speeds up the first pass for junior reviewers or business teams handling routine vendor agreements. The main limitations are that models can miss subtle risk that depends on broader business context, and they can occasionally flag standard language as risky or vice versa, so high-stakes contracts still warrant a qualified attorney's sign-off. Used as a triage layer, AI review lets legal teams spend their time on genuinely unusual contracts instead of reading every routine NDA line by line. Nanobase AI builds contract review pipelines against a customer's own playbook rather than a generic risk checklist.
The playbook is the system, not the model
An AI contract review tool is only as good as the playbook it compares clauses against, since the model's job is to match language against defined acceptable and unacceptable positions, not to independently decide what counts as risky. Building the playbook, explicit definitions of standard, negotiable and unacceptable positions for each clause category, is the majority of the engineering and legal effort in a contract review system, while prompting a model to apply that playbook consistently is comparatively straightforward. A generic "flag risky clauses" instruction without a defined playbook produces inconsistent, hard-to-trust output, because risk is inherently relative to a company's own standard positions, not an absolute property of a clause.
Legal teams that already maintain a negotiation playbook in some form have most of the hard work done; teams without one need to codify their standard positions before an AI review layer can be genuinely useful rather than just impressively fast.
Common clause categories and typical flags
| Clause category | Standard position example | Flag trigger |
|---|---|---|
| Indemnification | Mutual, capped at contract value | Uncapped or one-sided indemnification |
| Limitation of liability | Capped, carve-outs for gross negligence | No cap, or cap below acceptable floor |
| Termination rights | Termination for convenience with notice | No termination for convenience, or excessive notice period |
| Auto-renewal | Notice period defined, opt-out clear | Silent or short-notice auto-renewal |
| IP assignment | Clear ownership terms for deliverables | Ambiguous or missing IP assignment |
| Data protection | Aligned with applicable privacy law | Missing data processing terms |
Assigning severity so review time goes to what matters
Not every deviation from the standard position carries the same risk, and a system that flags every minor wording difference at the same severity trains reviewers to ignore flags entirely. A severity scheme, commonly high, medium and low, tied to concrete criteria, such as financial exposure or whether the deviation is negotiable versus a hard stop, lets a legal team triage a stack of contracts in minutes rather than reading every flag with equal attention. High-severity flags, like an uncapped indemnification clause on a large contract, should route for immediate attorney review, while low-severity wording variations can often be handled by a paralegal or accepted as within tolerance.
From flag to redline suggestion
The most useful contract review systems go beyond flagging a clause as risky and suggest a specific fallback position drawn from the same playbook, such as proposing a liability cap at a defined percentage of contract value when the original clause is uncapped. This turns the tool from a detection system into something that accelerates the actual negotiation, since a reviewer can accept, adjust or reject a concrete suggestion far faster than drafting a counter-position from scratch. Suggested redlines should always be reviewed before sending to a counterparty, since the model can occasionally propose language that does not fit the broader deal context.
Frequently asked questions
Does a contract review AI replace legal review entirely?
No, it accelerates the first pass by surfacing deviations and suggesting fallback language, but material contracts still warrant a qualified attorney's review, particularly for anything the model flags as high severity or anything falling outside the playbook's defined categories entirely.
How is the playbook kept current as company policy changes?
The playbook should be version-controlled and updated whenever legal or business leadership changes a standard position, with the review system referencing the current version so flags stay aligned with actual company policy rather than a stale, outdated one over time.
Can this work for contracts in a language other than English?
Yes, provided the underlying model has strong capability in that language and the playbook's standard positions are defined or translated accurately, since a mismatch between the playbook's language and the contract's language increases the risk of missed or misapplied flags.
How Nanobase AI helps
Nanobase AI builds contract review pipelines against a customer's own playbook, with severity scoring and fallback language suggestions, rather than a generic risk checklist that does not reflect the company's actual negotiation positions. See Nanobase AI's solutions for how this fits an existing legal workflow. Related: how reliable is AI contract analysis compared to a lawyer.
Ready to discuss your project? Contact Nanobase AI or email hello@bumu.tech.