AI in Biz Av Transactions

When a new client sent through 20 pages of AI suggestions for amendments to their Aircraft Purchase Agreement earlier this year, my eyes lit up. The potential promised by AI (such as efficiency, creativity and speed) is interesting, even to traditionally cautious lawyers. This situation gave me the perfect opportunity to test whether AI really could help negotiate complex business aviation transactions.

Our client had used three separate large language models (“LLMs”), including ChatGPT, to review the draft agreement and suggest improvements. The result was 77 proposed amendments. Of those:

• 6 worked and could be implemented directly;
• 18 were roughly in the right direction but needed tweaking to work in the agreement; and
• the remaining 53 were off the mark — they made the right kind of “noises” but just did not work in the transaction either commercially, legally or contextually.

Some of the issues with the suggestions will be familiar to anyone who has played with AI: confidently wrong, occasionally contradictory, and sometimes bewilderingly optimistic. A lot of the suggestions were based on sound legal principles and perfectly suitable for some form of contract, just not something as specific as a business aviation purchase agreement. Some suggestions were suitable in isolation but contradicted other parts of the draft agreement. Others were either based on commercial aviation practice or did not account for well-established market standards in business aviation.

The exercise may not have saved time on that deal, but it did highlight an important principle: AI can be an impressive tool, but it needs to be paired with human judgment. Users need to be aware that, in its current form and on the commonly used platforms like ChatGPT, the responses they get may lack context, nuance, or credibility. They must be checked.

The reason for this is structural. LLMs don’t “know” anything per se. They work by being trained on vast amounts of data to learn statistical patterns and then answer users’ queries by predicting “bites” or “chunks” of information based on that data. In each instance, the AI tool will predict the next most likely “chunk” given the relevant context. This can even lead to it seemingly “making up” answers, which is known as AI “hallucinations”. This may lead you to have some sympathy for those who are lulled into the trap of assuming that AI is correct.

In a case reported earlier this year (and there have been several similar cases), AI “hallucinated” false case law that was cited in legal proceedings. This could occur in instances where a user asks AI to conduct legal research and, in its response, it predicts that the next “chunk” of text should look like a case citation — a cluster of names, numbers, and law reports — without any awareness that it ought to check if such cases even exist. Without human intervention, AI does not understand why a lawyer might need a case or to know that no such case exists.

The High Court acknowledged that AI can be a useful tool. It stated that it will likely “have a continuing and important role in the conduct of litigation in the future”. It also stressed, however, that oversight is crucial. Lawyers have professional duties to the courts and to their clients, which include ensuring that their research, advice, and drafting is accurate.

This message also applies beyond the courtroom. AI use needs to be responsible and intentional. What could that use look like? For transactional business aviation lawyers, that may mean:

• feeding LLMs with a curated precedent bank so that their learning is focused and accurate rather than based on those very few business aviation articles that it can find online (articles from 2018 are great, but the Air Navigation Order has been updated since then!);
• building a platform that re-runs an entire draft through the model after each amendment so that the AI’s analysis stays contextual and it doesn’t lose sight of the agreement as a whole;
• embedding version-tracking and/or legislative-update checks into drafting workflows.

There’s real potential here. Used intelligently and intentionally, AI could speed up document review, flag inconsistencies, or draft first-pass content. But it can’t replace human expertise, especially in industries that depend on market understanding, precision, and/or safety.

Technology is extraordinary, and its future in law and business is assured. For now, though, our 77-amendment exercise proved that just like a jet: AI still needs a human pilot.