AI can draft the contract. It can't carry the liability.

In June 2023, a federal judge in New York sanctioned two lawyers and their firm $5,000 after they filed a brief citing judicial opinions that did not exist. The citations had been generated by ChatGPT. The court was explicit that using an AI tool was not itself improper. The failures included submitting unchecked output, continuing to defend the fabricated citations, and making false statements to the court.¹

The case became shorthand for AI risk in legal work, and it is often read as an argument against using AI at all. It is better read as an argument about process.

The tools are useful, and they still make mistakes

Purpose-built legal AI is a real improvement on general chatbots, but it is not error-free. In a preregistered study published in the Journal of Empirical Legal Studies in 2025, researchers at Stanford RegLab and the Stanford Institute for Human-Centered AI tested leading legal research tools on more than 200 legal queries. Lexis+ AI produced hallucinated information about 17% of the time, and Westlaw's AI-Assisted Research did so about 33% of the time.² The authors concluded that vendors' claims of hallucination-free research were overstated.

Professional guidance has settled on the same point. In July 2024, the American Bar Association issued Formal Opinion 512, its first formal ethics opinion on generative AI. It confirms that lawyers may use these tools, and that the duties of competence, confidentiality, and supervision apply to everything the tools produce.³

Where AI belongs in legal operations

Most legal work inside a business is not litigation. It is intake, review, tracking, and preparation, and that is where AI helps most:

• Triage. Sorting incoming contracts and requests by type, risk, and urgency.
• First-pass review. Comparing a draft against the firm's standard positions and flagging every deviation.
• Claims checking. Confirming that marketing and sales statements match approved, sourced claims.
• Obligation tracking. Extracting dates, renewals, and commitments into a record someone owns.
• Counsel preparation. Assembling the facts, documents, and open questions so an attorney's time goes to judgment rather than gathering.

The line that matters

In every one of these, the system prepares and a qualified person decides. That line has to be designed, not assumed. The workflow should state what the system may clear on its own, what it must escalate, and who signs off. Every output a person relies on should carry its sources, so verification takes minutes rather than hours.

Auxeon's legal operations specialist, Javi, is built on that boundary. Javi checks, maps, and prepares, and attorneys advise. Javi provides operational support, not legal representation or legal advice, and the design makes that visible at every step.