Last month, two lawyers filed a petition in a Brazilian labor court that contained a secret message aimed at tricking the court’s AI tool into favoring their client. This type of AI mischief, known as a “prompt injection,” was written in white text on a white background and thus invisible to humans. Translated from Portuguese, the prompt injection read as follows:

‍The scheme was designed to tamper with the court’s AI to win the case. It failed because the court’s AI detected this effort. The judge was alerted and then sanctioned the two lawyers.

Unfortunately, there is no silver bullet today to defeat all prompt injections. Prompt injections take advantage of a basic limitation in how AI currently works: it has difficulty telling the difference between text it should read and instructions it should follow. Hidden text can be planted in a brief, a contract or an opposing party’s discovery production. Any document or text can serve as a “Trojan horse.” Indeed, prompt injections improve in a way upon the original Trojan horse: if the invasion succeeds, no one ever finds out it happened.

For over a year, the legal community has focused on a different type of AI mischief known as “case hallucinations.” The AI invents out of whole cloth case citations that appear to support the lawyer’s position; the lawyer then inadvertently relies on these fictional citations in a memo or brief. Or the AI may identify an actual case but quote or characterize it inaccurately, unfairly or out of context. Some of these issues can be solved with a thorough, old-fashioned cite check, where a human lawyer actually reads the case to confirm that it soundly supports the client’s position and cannot be used by opposing counsel to damage the client’s position and undermine the lawyer’s credibility.

As the cautionary tales of prompt injections and case hallucinations make clear, AI is not only a source of efficiency, it generates new risks that lawyers will be called upon to manage. AI technology is incredibly complex and becoming more so. Many lawyers are not technologically proficient let alone in AI. Nonetheless, we owe professional obligations to clients and to the legal system in an ever more demanding technological environment.

One partial solution is to try to develop a layman’s understanding of the potential threats and the tools available to counter them. Another is to retain experts who have the requisite knowledge and can make recommendations as to best practices. In all events, we as lawyers can proactively think through how best to structure our own work, and supervise others, to manage AI risks. Ignoring those risks is no defense.

---

This publication is for general information only, does not create an attorney-client relationship and should not be relied upon as legal advice. © 2026 Gruenstein Law PLLC.

‍