Why "I Didn't Know It Was AI" Is Not a Defense

In every sanctions case involving AI-generated citations, the attorney has offered some version of the same defense: I did not know the AI would fabricate citations. I trusted the tool. I did not realize it could make things up. Courts have uniformly rejected this defense.

The reason is simple. Rule 11 of the Federal Rules of Civil Procedure requires that every filing be supported by an attorney's certification that "the claims, defenses, and other legal contentions are warranted by existing law." This obligation predates AI by decades. It does not contain an exception for citations generated by technology the attorney does not fully understand.

The attorney's duty to verify is not delegable. Whether a citation came from a law clerk, a contract researcher, a Westlaw search, or a ChatGPT prompt, the signing attorney is responsible for its accuracy. The tool is irrelevant. The obligation attaches to the signature.

Some attorneys have argued that AI tools are analogous to legal research databases, and that attorneys have never been expected to independently verify every result from Westlaw or LexisNexis. But this analogy fails on a critical point: research databases return results that exist. They may be incomplete or out of date, but they do not fabricate cases. AI language models can and do.

The practical implication is that AI tools should be treated as drafting assistants, not as research tools. An attorney who receives a draft from a junior associate would never file it without checking the citations. The same standard applies to a draft from an AI tool.

Going forward, the standard of care is clear. If you sign a filing, you are certifying that every citation in it points to a real authority. The method of verification is up to you. But the obligation to verify is not optional, and ignorance of your tools is not a defense.