An Indonesian exporter ran a €50k supply contract through ChatGPT. The generated terms looked complete: payment terms, delivery, dispute resolution. Six months in, a goods-in-transit delay triggered a liability dispute. The contract had no force majeure clause. It also had no mention of NPWP (Nomor Pokok Wajib Pajak—the tax ID required to legally invoice in Indonesia). When the supplier tried to issue an invoice, it was rejected outright. The contract was legally sound in a generic sense. It was worthless in practice. This is not an edge case. AI language models train on English-language contracts and generic legal templates. They optimize for fluency and completeness-on-paper. They do not optimize for regional law, contextual specificity, or the gaps that actually sink deals. If you're using AI to draft contracts—especially across Southeast Asia—you're accepting legal and financial risk that you probably don't realize you're taking. What AI contract drafters systematically omit Large language models are pattern-matching engines. They work by predicting the next most-likely token based on training data. For contracts, that training data is overwhelmingly Anglo-American, English-language, and published. What gets omitted is regional, contextual, and not in the training set. Missing regional clauses Malaysia requires SSTIN (Service and Sales Tax Identification Number) disclosure for service contracts above certain thresholds. Indonesia requires NPWP matching on every invoice. Singapore requires entity registration details and GST treatment confirmation on B2B contracts. AI drafts none of these by default. It will mention 'tax compliance' in generic terms. It will not specify which tax ID, where it goes, what triggers validation, or what happens if it's missing. A contract that omits these details may be signed and still unenforceable the moment payment is due. The invoice gets rejected. The buyer withholds payment pending 'correct' documentation. You're now in a 30–60 day dispute over something that should have been a two-line clause. Liability caps that don't match context AI tends to suggest liability caps that are symmetric: each party's liability is capped at the total contract value, or sometimes at annual fees. This works for commodity SaaS. It breaks for supply contracts, services with embedded IP, and deals where one party's breach cascades into client liability. If you're a reseller and your supplier's product fails, you may owe your customer damages that exceed the supply contract value by 5–10x. A symmetric cap leaves you exposed. A lawyer would build in tiered caps: supplier caps at contract value, your cap at direct damages only, client liability exclusions on both sides. AI will not generate this level of nuance without explicit instruction—and most users don't know to ask for it. Tax treatment assumptions that are wrong Goods-and-services contracts in Southeast Asia carry tax implications that are not globally uniform. Services in Malaysia trigger SST (Service and Sales Tax) at 6%. Goods imports into Indonesia trigger import duty and VAT stacking. E-services (SaaS, consulting delivered remotely) may be exempt, reverse-charged, or GST-liable depending on the supplier's residency and the buyer's status. AI drafts generic 'tax is the responsibility of each party' language. This does not specify whose tax ID applies, whether GST/VAT applies, whether it's inclusive or exclusive of price, or what happens if tax treatment changes mid-contract. A supplier in this position discovers mid-billing that their invoicing software won't accept the contract terms. A week of back-and-forth follows. Payment is delayed another two weeks pending re-invoicing. Dispute resolution that ignores local law AI often suggests international arbitration and English law as dispute resolution. This is the default for large, English-language contract templates. For a SMB contract between an Indonesian exporter and a Malaysian importer, international arbitration in Singapore costs ₹50–100 lakh just for filing and hearing. English law is a foreign jurisdiction that neither party knows intimately. A contract that calls for this makes small disputes uneconomical to pursue—which is not a feature, it's a flaw. Local courts, local law, and a narrowed arbitration scope (confined to payment disputes, for example) would be more practical and more enforceable. AI does not reach for practical by default. How to audit an AI-drafted contract This does not mean never use AI to draft contracts. It means AI is the first draft, not the final one. You need a legal review process. Here's what that looks like: Tax and compliance sweep Is the supplier's tax ID explicitly named? (e.g., 'Supplier's NPWP: [___]' for Indonesia, or 'Supplier's ABN: [___]' for Australia). If not, add it. Does the contract specify whose tax obligation applies for each line item? (goods, services, e-services, delivery). If it says 'each party bears its own taxes', you've de