The judge bears full responsibility for every word in a final document, no matter how a draft was generated. Failure to review GenAI-generated work product carries significant risks to the just adjudication of disputes (and the judge’s reputation).

Using GenAI to develop the first draft of a judicial opinion may inappropriately abdicate independent analysis and bias subsequent drafts in the direction of the AI’s reasoning. Using GenAI to draft remarks introducing a speaker at a conference must be done with care; GenAI has been known to cite inaccurate biographical information (because it is designed to predict plausible language not verify facts). However, using GenAI to draft an email or compose a conference presentation from a series of self-generated bullet points is relatively low risk (though technical content should always be verified).

Relying on GenAI to draft an opinion (or other types of documents like scholarly articles) can implicate cognitive biases such as automation and confirmation bias. When GenAI is used to create a first draft, there is a tendency to accept the output as accurate and objective. In addition, writers tend to get ‘anchored’ to first drafts and work from them, even if they are not optimal, rather than starting afresh. For this reason, it may be advisable to use GenAI to review an outline or refine a draft rather than to initiate the writing process. 

GenAI output can sound generic. If using GenAI to generate or revise drafts, judges can train the GenAI system to produce text more closely resembling their style and voice by uploading a collection of prior rulings and other writings. Judges also can customize settings for GenAI drafting including avoiding the use of contractions or em dashes.

GenAI can assist with preparing general remarks for public speaking, drafting internal memos to staff, editing for grammar and clarity, and creating templates for routine procedural orders. It may also be used to ‘check one’s work.’ For example, after completing a draft of a document, prompts can be developed to test the strength of an argument or identify possible gaps in reasoning. (Note: the Codes of Conduct Committee AI guidance considers draft opinions to be confidential information and advises against uploading into a GenAI system.)

Using GenAI to assist with drafting initial versions of complex orders, bench memos, or opinions requires great caution. This type of use may compromise critical thinking and judgment. For example, because the initial GenAI-assisted draft will necessarily choose a logical framework to sequence, organize, and reason through the issues, it may be harder for the author to see alternative approaches. In addition, it can be more difficult (and more time consuming) to detect and correct flawed legal reasoning in the editing process than it is to draft correct legal reasoning in the first place.

Chambers staff should always disclose their use of GenAI to the judge, including what tools they used and how they were used. Whether there needs to be disclosure to the parties or outside counsel is a harder question. Judges must independently determine which use cases, if any, warrant disclosure. Section D of the Codes of Conduct Committee’s ethics guidance states: 'Depending on how AI tools are used in the adjudication process, judges must consider whether parties should have the opportunity to respond to AI-generated information used by the court.' If a judge’s GenAI research goes outside the record, it may be considered ex parte, warranting disclosure (as well as an opportunity for parties to respond).