Odette CaplanWhen OpenAI launched GPT-6 on October 7, 2026, for eligible ChatGPT Business workspaces, it triggered a wave of adoption across U.S. small businesses, startups, and independent contractors. The promise was irresistible: a contract draft in seconds, no billable hours, no waiting for a lawyer to call back. But legal professionals are raising a quiet alarm — and the question they're asking isn't whether GPT-6 can write a contract. It's: who's on the hook when it writes one badly?
The GPT-6 Contract Moment
GPT-6's October 7 rollout paired the model with a new Intelligent UI, allowing businesses to generate interactive documents, forms, and contracts directly within ChatGPT Business workspaces. Access expanded rapidly to millions of U.S. businesses in the weeks that followed. For many, the immediate use case was paperwork: NDAs, service agreements, freelance contracts, vendor terms.
The appeal makes sense. GPT-6 can produce a five-page service agreement in under two minutes, complete with indemnity clauses, limitation of liability sections, force majeure provisions, and governing law language. For a cash-strapped founder or a solo consultant navigating a new client relationship, the time and cost savings appear significant.
But a January 2026 article in the Southern California Law Review titled "Artificial Incompetence?" found that large language models still struggle with the nuance that determines whether a clause is enforceable. The research identified four recurring failure modes: hallucinated legal citations, inconsistent cross-referenced clauses, outdated statutory references, and invisible ambiguities that only surface in litigation.
The AI-drafted contracts that cause the most damage, the paper noted, don't look wrong. They look like contracts.
Why Legal Stakes Are Higher Than You Think
Here is the uncomfortable truth about AI contract drafting: the contract is legally binding on you, not on the model.
No court in the United States has allowed a party to void a contract because an AI tool drafted it poorly. Courts hold AI-generated contracts to exactly the same enforceability standards as handwritten ones — offer, acceptance, consideration, mutual assent. If your GPT-6-drafted service agreement contains a clause that inadvertently waives your right to seek consequential damages, that waiver stands. If it references a statute that was amended two years ago, the outdated version you relied on offers you no protection.
The FTC's guidance on AI in business practices makes clear that deploying AI tools in commercial transactions does not shift liability away from the business that used them. The FTC has specifically flagged AI-generated contracts in B2C contexts as an emerging area of consumer harm risk, particularly where consumers may not know their agreements were produced without human legal review.
Legal malpractice insurers have taken note of the professional liability angle. A February 2026 CLE course from BARBRI, aimed at U.S. attorneys, warned that lawyers who rely on AI-drafted contracts without line-by-line human review face exposure for breach of their professional duty of competence — even if the final document reads professionally and "looks" correct.
The trend is also being tracked by courts in a related context. As detailed in coverage of the Sam Altman and OpenAI vendor risk litigation, U.S. businesses that embedded AI tools in core operations without adequate legal review have faced unexpected liability when those tools performed differently than advertised. The same principle extends to AI-drafted contracts: the business that deployed the tool bears the consequences of its output.
What GPT-6 Actually Gets Wrong
The errors in AI-drafted contracts are rarely obvious. GPT-6 does not write contracts full of typos or visible nonsense. It writes contracts that read professionally and sound authoritative. The problems emerge in the fine print.
Hallucinated statutes. The model may cite a specific section of the Uniform Commercial Code or a state consumer protection law that doesn't exist, or that was amended before its training data reflects the change.
Mismatched indemnity scope. In multi-clause documents, GPT-6 may write an indemnity provision in Section 7 that contradicts a carve-out in Section 12. Neither clause is wrong in isolation — together, they create an ambiguity that courts resolve against the drafter. That's you.
Outdated governing law. In employment contracts, the model may apply pre-2024 non-compete standards in states like California, Minnesota, or Oklahoma — states that have banned or severely restricted non-compete clauses. A GPT-6-drafted non-compete clause in those jurisdictions isn't just unenforceable; it can expose the drafting party to a counterclaim.
Missing jurisdiction-specific protections. State consumer protection laws, local landlord-tenant statutes, and industry-specific regulations (healthcare, financial services, real estate) change frequently. GPT-6 cannot reliably stay current, and it will not flag when a clause violates a newly enacted state law.
Entertainment lawyers have already documented a version of this problem. In the wave of AI-related contract disputes covered following Gal Gadot's six-month AI contract battle, attorneys noted that AI tools consistently failed to account for jurisdiction-specific performance rights language — the same class of error now appearing in business contracts.
When $8,500 Became a $43,000 Problem: A Concrete Case
Consider a scenario that mirrors what attorneys in several states are now reporting as a pattern.
A freelance software developer based in Dallas, Texas used GPT-6 in October 2026 to draft a $8,500 fixed-fee contract with a startup client. The contract included a standard limitation of liability clause capping the developer's exposure at the contract value. The clause looked correct and read confidently.
What the developer did not notice: GPT-6's clause excluded "gross negligence" from the cap — a standard carve-out — but also used language that Texas case law could read to exclude any claim from a "fundamental breach." Under Texas precedent, a missed deadline that derails a client's product launch can qualify as exactly that.
The project delivered three weeks late. The startup claimed $43,000 in downstream losses — a lost pre-order window, emergency contractor costs, and a delayed Series A milestone — arguing the delay constituted a fundamental breach. The developer's GPT-6-drafted liability cap provided no protection.
If the developer had paid a Texas contracts attorney $450 to review the agreement before signing, the liability cap would have been enforceable and damages would have been capped at $8,500. Instead, she faced a $43,000 claim plus $9,000 in legal fees to defend it.
The if/then logic is clear: if you use GPT-6 to draft a contract for work valued over $5,000, and a licensed attorney does not review it before signing, you are accepting full legal responsibility for every clause the AI generated — including the ones that worked against you.
What You Should Actually Do
Using GPT-6 to draft a first version of a contract is not inherently dangerous. The tool is fast, reasonably comprehensive for common scenarios, and a useful starting point. The risk is treating a first draft as a final document.
Legal professionals recommend a tiered review approach based on contract value and complexity:
- Under $1,000: AI drafting with a structured checklist review is adequate for simple services between parties with an existing relationship and a clear scope of work.
- $1,000 – $10,000: A lawyer's flat-fee contract review (typically $350–650 for a one-hour review) is strongly advisable. Focus review on indemnity, liability cap, and governing law clauses.
- Over $10,000: Full attorney drafting or a comprehensive clause-by-clause review is non-negotiable. Disputes at this level are expensive enough that prevention almost always costs less than litigation.
For contracts involving employees, software IP ownership, equity agreements, real estate, or any regulated industry, no contract value threshold applies — a licensed attorney is required regardless of how confident the AI-generated draft appears.
The businesses that are navigating GPT-6's contract capabilities most successfully are the ones treating the model as a drafting assistant — generating speed and structure — while relying on experienced legal counsel to verify enforceability and catch what the model missed.
If you're working with contracts that carry real financial or legal risk, connecting with an experienced contracts attorney on ExpertZoom can help you identify exactly where your GPT-6 draft falls short — before the other side's counsel does.
YMYL Disclaimer: This article is for informational purposes only and does not constitute legal advice. Contract law varies by state and by the specific facts of each situation. Consult a licensed attorney before signing or relying on any contract.
