The Post-Avianca Scrutiny of AI: A New Standard for NJ E-Discovery and Your Duty Under RPC 3.4

The legal profession is still processing the fallout from cases like Mata v. Avianca, where AI-generated legal research led to significant court sanctions. While the headlines focused on fabricated citations, the core lesson for New Jersey attorneys is far broader: you are unequivocally responsible for the output of your AI tools. This principle applies with even greater force in the high-stakes, high-volume world of e-discovery, where a tool’s error isn’t just embarrassing—it can directly impact your duties to opposing counsel.

For solo practitioners and small firms in New Jersey, Technology-Assisted Review (TAR) is no longer a luxury; it’s a necessity for handling document-intensive cases. Yet, simply licensing a tool and clicking “run” on a production is a dangerous oversimplification. It overlooks a crucial ethical obligation under NJ RPC 3.4: Fairness to Opposing Party and Counsel.

Beyond Competence: The Fairness Obligation of AI-Assisted Review

Most discussions about AI and ethics revolve around RPC 1.1 (Competence). While essential, focusing solely on competence misses the point in discovery. RPC 3.4(a) states that a lawyer shall not “unlawfully obstruct another party's access to evidence.”

Consider this scenario: you use an AI tool to review 100,000 client documents. Due to a poorly constructed seed set or an opaque algorithm you don’t fully understand, the tool incorrectly flags 5,000 responsive documents as non-responsive. You produce the remainder, signing a discovery response stating your production is complete. Have you just violated RPC 3.4?

A compelling argument can be made that you have. A reckless or negligent failure to validate your AI’s process isn't a mere technical glitch; it's a failure that actively obstructs the other side's access to discoverable information. The excuse that “the AI did it” will offer no protection from a motion for sanctions or a challenge to the defensibility of your review process.

Four Practical Steps to Bulletproof Your AI Discovery Process

To meet your obligations under RPC 3.4 and build a defensible AI workflow, New Jersey firms must move from being passive users to active supervisors of their technology. Here’s how:

1. Demand Process Transparency from Your Vendor

Before you sign a contract, ask vendors pointed questions about their algorithms. How does the TAR model prioritize documents? What factors does it weigh? Can you audit its coding decisions? If a vendor’s answer is simply “it’s a proprietary black box,” that is a major red flag. Your duty of competence requires you to have a fundamental understanding of the tool’s mechanics to properly supervise it. You must be able to explain and defend your process to a court.

2. Implement and Document a Rigorous QC Protocol

Trust, but verify. The gold standard for defensible TAR is a robust quality control (QC) sampling process. This means manually reviewing a statistically significant, random sample of documents that the AI has categorized. Critically, this includes sampling from the documents the AI marked as non-responsive. This is where smoking guns are often missed. Document every step of your sampling methodology—the sample size, the error rate you discover, and any adjustments you make to the AI model as a result.

3. Proactively Negotiate AI-Specific Clawback Terms

Standard clawback agreements are a good start, but they were designed for manual human error, not algorithmic error. When you know you are using TAR, be proactive. Propose discovery protocol language or a clawback provision that specifically acknowledges the use of AI-assisted review and establishes a clear, cooperative framework for rectifying the inadvertent production of privileged material by either party's AI system.

4. Train Your Team on the Why, Not Just the How

Your paralegals and junior associates are your first line of defense. It’s not enough to train them on which buttons to click. They must understand why the QC sampling protocol is ethically necessary. They need to be empowered to flag anomalies in an AI’s output and understand that their careful oversight is directly tied to the firm’s ethical obligations under the Rules of Professional Conduct. This is a direct extension of a partner’s supervisory duties under RPC 5.1 and 5.3.

The scrutiny of AI in legal practice is only beginning. For New Jersey litigators, the focus must expand from the risks of generative AI in research to the procedural integrity of AI in discovery. By centering your process on the duty of fairness enshrined in RPC 3.4, you not only protect yourself from sanctions but also build a more robust, reliable, and ultimately more just discovery process.