AI contract generation tools are genuinely useful for speed. A template-heavy NDA, employment agreement, or service contract that would take a lawyer 2–3 hours to sketch can be drafted in minutes. But "fast" is not "correct," and the gap between them is where your business gets exposed. The problem is structural. AI learns from training data—patterns in contracts that exist, not rules that should exist. It's excellent at replicating common clauses. It's dangerous at handling edge cases, jurisdiction-specific law, and the precise financial terms that actually protect you. A junior lawyer reviews AI-generated contracts and finds gaps immediately. You don't always have that junior lawyer on speed dial. This is especially true if you operate across Southeast Asia. Malaysia has different employment law than Singapore, which differs from Indonesia. AI doesn't know the difference unless you tell it explicitly—and most general-purpose AI tools don't ask. The four gaps AI consistently leaves open 1. Jurisdiction-specific mandatory clauses Employment contracts in Malaysia require explicit reference to statutory benefits: annual leave, sick leave, and the conditions under which notice can be given. Singapore employment law mandates specific language around misconduct and termination procedures. Indonesia's labor code requires clear job descriptions and probation period rules inside the contract itself—you can't handle these separately. A generic AI contract template won't mention these unless you prompt it directly. And if you prompt it, the AI often adds boilerplate that doesn't align with how these laws actually work in practice. It might say "severance in accordance with local law," which is legally correct but tells you nothing about quantum or calculation method, leaving your finance team guessing later. 2. Payment term gaps and currency hedging If you invoice across borders—Malaysia in MYR, Singapore in SGD, Indonesia in IDR—your contract needs to specify who bears currency risk. AI drafts payment terms generically: "Invoice due within 30 days." It rarely specifies whether prices are fixed in the invoice currency, whether exchange rate movements are the client's problem or yours, or how you handle payment delays that cross month-end (when rates shift). For retainer or milestone-based contracts, AI often misses the question of when payment is actually due relative to delivery. Is payment 30 days from invoice, 30 days from delivery, or 30 days from client sign-off? The AI might say "30 days" without linking it to an event. Six months later, your client says they haven't "accepted" delivery so the 30-day clock hasn't started. You're not legally wrong—you're just not paid. 3. Approval and rejection workflows inside contracts If you're selling a service that requires client approval (design, copy, development), your contract needs to define what "approval" means. How long does the client have to review? What counts as "approved"? What happens if they reject work—do they pay for time spent, or is it your loss? AI drafts vague language: "Client shall approve deliverables in writing." It doesn't define the timeline, the rejection process, or revision limits. This is where most contract disputes start. Not breach of core terms, but disagreement over what the contract actually requires when things go sideways. AI doesn't anticipate sideways. 4. Liability caps and indemnification mismatches AI almost always includes liability caps—often a blanket "liability limited to fees paid in the last 12 months." This sounds protective until you read the indemnification clause, which often carves out exceptions without actually saying what those exceptions are. So you're capped at 12 months of fees in damages—except when you're not, because the indemnification is unlimited. The two clauses contradict each other. Even worse: if you're serving clients in regulated industries (financial services in Singapore, healthcare in Malaysia), AI doesn't know whether indemnification caps are even legal. Some regulators require liability to be uncapped for certain harms. Why this matters more in Southeast Asia Malaysia, Singapore, and Indonesia have different legal systems. Malaysia and Singapore inherited English common law, but their statutes diverge—Singapore has moved faster on data protection and employment law modernization. Indonesia uses civil law, with labor law heavily weighted toward employee protection. AI trained on English contracts and US precedent doesn't know these systems exist. More pragmatically: if you're hiring across all three countries, your employment contract templates need to be jurisdiction-specific. A generic "employment agreement" might be legal in Singapore but insufficient in Malaysia (missing mandatory clauses) and illegal in Indonesia (violating minimum wage or probation rules). AI won't flag this. It will happily generate a contract that sounds professional and is actually non-compliant. Build a template audit che