Delivery and mobility / Ride-hailing and delivery
Grab: Pilot automating the Legal team's first-pass review of non-disclosure agreements (NDAs) with AI extraction and rule-based decisions
- Company
- Grab
- Country
- Singapore
- Adoption stage
- Pilot
- Source published
- Date basis
- The date the source was published. It can differ from the date adoption started.
- How the source was checked
- Read the full source text
The work problem
According to an article Grab published on May 4, 2026, its Legal team repeatedly carries out the first-pass review of non-disclosure agreements to check whether a contract follows Grab's standard terms or needs a lawyer's closer attention. In a first experiment in which a single large language model (LLM) read an NDA and decided whether it should pass, be flagged or be escalated, the model could not apply policy checks consistently across a full contract. It sometimes missed non-standard wording, interpreted the same phrases differently across runs and produced recommendations that sounded plausible but did not stand up to an audit-grade review.
Technology and data
Grab's Legal team and Grabber Technology Solutions, its internal corporate IT department, built a three-layer decision engine in which AI reads, code decides and humans approve. Multiple LLMs extract facts and exact citations from the contract, and a consensus vote across models stabilizes the outputs. A deterministic rulebook written and owned by the Legal team checks the extracted facts against Grab's policies and decides whether to auto-approve, suggest a standard amendment or escalate to Legal. A reviewer interface shows facts, citations and rule outcomes side by side, and anything the AI cannot decide automatically is escalated to legal experts.
Results
Grab said that when it tested the engine against hundreds of contract variants spanning compliant, low-risk and high-risk scenarios, it achieved 100% rule evaluation accuracy and 0% false positives, and review time was cut from 60 minutes to under 2 minutes. The Legal team retained the final say at every step. The company plans to keep stress-testing its risk management framework, then move the NDA engine from the sandbox to production and expand it to commercial contracts.
Limits and open questions
The results were measured on contract variants in a sandbox, and the number of contracts handled in production has not yet been disclosed.
Sources
Compiled from public sources. These are not results from ATF Works customers.