You run a seven-person team across Malaysia and Singapore. A contractor in Kuala Lumpur needs an agreement by Friday. You open ChatGPT, paste in a template, and ask it to customize the payment terms. Thirty seconds later you have a draft. You send it to your lawyer on Monday morning. By Tuesday, she's circled seven gaps—some of them serious enough to void the whole thing if they ever hit a Malaysian court. This is not a worst-case scenario. It happens every week across Southeast Asia. The problem is not that LLMs are useless for contracts. It's that they are plausibly fluent at writing something that reads like a contract, feels professional, and contains enough legal language to look legitimate. Then it fails silently in ways that cost money when it matters. Here are the seven most common liability gaps that AI drafts into contracts, and why they survive initial review until a lawyer or a court catches them. 1. Invented pricing and currency baseline An LLM will confidently write: "Service pricing is USD 5,000 per month, payable in Malaysian Ringgit at the prevailing exchange rate on the invoice date." Neither party mentioned USD. Neither party said anything about which exchange rate (mid-market, interbank, card rate?). The LLM inferred a currency and an index that do not exist in the source material—and that create ambiguity about what "prevailing" means when the Malaysian Ringgit moves 2–3% in a week. A better version: "Service pricing is MYR 22,000 per month, invoiced on the first business day of each month, due net-30 to account [number]." No inference. No phantom currencies. No floating rate that triggers disputes. LLMs do this because they are trained to be helpful and to fill gaps. When you do not specify a currency, the model reasons: "Professional contracts include currency." So it picks one. When you do not specify an exchange rate rule, it reasons: "Exchange rates are important." So it invents a mechanism. Both inferences are wrong because contracts must be explicit , not inferred. 2. Flattened jurisdiction and governing law clauses AI drafts: "This agreement shall be governed by the laws of Malaysia and disputes resolved by arbitration in Singapore." Sounds professional. Creates chaos in court. Which laws apply to interpretation? Which court has jurisdiction if one party sues before arbitration begins? What if Malaysian tax law contradicts Singapore dispute law? The clause is so vague that it does not actually govern anything. A working clause requires nesting: Substantive law: "This agreement is governed by the laws of Malaysia, without regard to conflict of laws principles." Dispute escalation: "Disputes shall first be resolved by good-faith negotiation between [name] and [name] for 30 days." Arbitration mechanics: "If negotiation fails, disputes shall be arbitrated under the Rules of the [specific body, e.g., SICC], with a single arbitrator, seated in [city], with proceedings in English." LLMs flatten this into vague clauses because they do not understand that jurisdiction is hierarchical. They see "Malaysia" and "Singapore" and "arbitration" and jam them together into a single sentence that sounds authoritative but binds nothing. 3. Missing stamp duty and filing requirements A contract drafted by an LLM for a Malaysian service agreement will skip the stamp duty schedule entirely. Malaysia's Real Property Gains Tax Act and Stamp Act require many contract types to be stamped within 30 days of execution. If the document is not stamped, it is inadmissible in court. An LLM does not know this because stamp duty is a jurisdictional fact that appears nowhere in generic contract templates. The same problem hits Singapore (which requires certification for certain contracts) and Indonesia (which has specific requirements for contractor agreements and partnership documents). An AI that was trained on English-language corporate templates will have never seen these filing requirements, so it omits them. Your lawyer catches this on the first read. A regulator catches it on an audit. 4. Liability caps that invert the risk allocation LLMs love symmetry. So they draft: "Neither party shall be liable for indirect damages, and total liability shall not exceed the fees paid in the prior 12 months." Sounds balanced. Actually favors the service provider in every case where the customer suffers real loss. If you are the customer and your contractor's mistake costs you MYR 500,000 in lost revenue, you can recover only the fees you paid (maybe MYR 50,000). If you are the contractor and the customer's breach costs you MYR 500,000, you can also recover only fees paid. Symmetry is almost never what you want in contracts. Service providers want low caps. Customers want unlimited liability for core obligations. AI drafts symmetry because it sounds fair on paper. It is unfair in execution. 5. Indemnity clauses that do not actually protect either party AI drafts: "Each party shall indemnify the other against third-party cla