Lessons from Michael Cohen v. United States: Criminal Defendants Should Not Be at the Mercy of Technology for Privilege Review

New York Law Journal

Michael Cohen’s months-long legal battle captivated the public. With hush money payments to strippers and the National Enquirer as well as presidential tweet storms, it’s no surprise Cohen’s Dec. 12, 2018 sentencing was front-page news. But Cohen’s criminal saga arguably began with an issue dreaded by attorneys and judges alike: the potential disclosure of attorney-client privileged communications.

On April 9, 2018, the FBI raided the office, home, and hotel room of President Donald Trump’s personal attorney, Michael Cohen, seizing business records, emails, and electronic data. Because Cohen is an attorney, the documents seized needed to be reviewed for attorney-client privilege before they could be turned over to prosecutors.

Privilege review options were the subject of much debate. The court considered using Technology Assisted Review (TAR) to expedite the privilege review process, but ultimately—and correctly—decided against it. While TAR is a common and beneficial tool in civil litigations, it is improper and potentially unconstitutional as the sole arbiter for privilege review in criminal cases, particularly if required by courts.

Technology Assisted Review

Technology Assisted Review or TAR is a document review process where humans train software to identify potentially relevant documents. Since Magistrate Judge Peck’s 2012 opinion in Da Silva Moore v. Publicis Groupe & MSL Group, 287 F.R.D. 182 (S.D.N.Y. 2012), approving the use of TAR, many litigants—particularly those faced with large data sets—have embraced TAR for its efficiency and cost savings.

TAR technology has evolved considerably over the past six years. In earlier software programs, attorneys would review and code a “seed set” of data. The software identified certain properties about the seed set and used what it “learned” to code other data. Currently, there are software vendors working on tools where the learning and validation process is already built in. For example, you could apply a predefined “gender discrimination” training module to your client’s documents rather than training the software yourself.

The Debate: How to Review Cohen’s Documents for Privilege

Cohen wanted either his attorneys or a special master to review the seized documents for privilege. The federal prosecutors initially objected to the appointment of a special master, insisting that a “filter team”—a group of prosecutors and investigators who are not involved in the case but may work in the same office as those prosecuting the matter—review the seized materials for privilege.

On April 26, 2018, however, prosecutors withdrew their objection to the appointment of a special master and offered a “compromise position”—that a Special Master be appointed to conduct the privilege review using TAR. In support of this proposal, prosecutors provided a letter from retired Magistrate Judge Frank Maas, one of the government’s proposed candidates for Special Master. Judge Maas’ proposal described how a TAR process known as continuous active learning could be used to review the seized material and make privilege determinations.

Judge Maas’ proposal touted TAR’s speed and efficiency, suggesting that “the entire technology-assisted review of the electronic data could likely be accomplished in one to two weeks, and would not require the use of contract reviewers or associates.” Judge Maas further wrote that, while he “recognize[d] the importance of this undertaking,” cost was a significant issue: “whether the Government or a master performs the initial review … the process of identifying those that are privileged and determining whether any privilege exceptions apply should not be inordinately expensive or time consuming.”

On April 26, 2018, the same day the government filed its letter advocating TAR, Judge Wood appointed the Hon. Barbara S. Jones as Special Master. At the hearing, Judge Wood noted that Judge Jones was familiar with TAR and was “technologically well suited to the job,” but left it to Judge Jones to decide how to conduct the privilege review. Ultimately, TAR was not required.

Rather, between May and July 2018, the government produced the seized documents to Cohen and certain parties who moved to intervene. Cohen and the intervenors reviewed the documents, identifying for the Special Master any documents they claimed contained privileged or highly personal information.

The Special Master reviewed the documents identified as privileged, frequently conferring with the parties. In the end, Cohen and the intervenors withdrew their privilege claims on over 2,000 documents with the Special Master recommending that all or part of approximately 7,400 documents were either privileged or highly personal and releasing over 1.3 million documents to the government. Neither Cohen nor the intervenors objected to the Special Master’s final recommendation.

While TAR was ultimately not used in the Cohen matter, the prosecution’s proposal is cautionary: TAR has been used to assist with privilege review, but for good reason has not been generally accepted as the sole arbiter of privilege.

The Case Against TAR for Privilege Review in Criminal Cases

TAR is rarely used for privilege. TAR is most often used for relevance determination but is not used in every case and is almost unheard of as the sole tool for privilege review. In an interview with Legaltech News, Judge Maas conceded that “using TAR for privilege review is more challenging than using it to find relevant documents.” Given these admitted challenges, prosecutors and courts should be even more restrained in proposing or ordering technology for such untested applications.

Defendants should make determinations regarding their documents
. Even civil litigants cannot be forced to use TAR. See, e.g., Rio Tinto PLC v. Vale S.A., 306 F.R.D. 125, 127 (S.D.N.Y. 2015) (denying requesting parties’ attempt to force producing parties to use TAR). While TAR has been used in privilege reviews, it has assisted human review—not replaced it. See Good v. Am. Water Works Co., 2014 WL 5486827 (S.D. W. Va. 2014). In Good, plaintiffs wanted defendants to only use TAR for the privilege review, entirely replacing human review, but the court sided with defendants and allowed both human review and TAR.

Moreover, not only is the responding party “best situated to evaluate the procedures, methodologies, and technologies appropriate for preserving and producing their own electronically stored information” as outlined in Sedona Principle Six, but the producing party should have the opportunity to review its own documents before the prosecution. In the context of a criminal prosecution, it is about the fundamental fairness of permitting a defendant the first opportunity to review his own documents for privilege. Indeed, in situations like Cohen’s—where a warrant is executed on an attorney’s office—the U.S. Attorney’s Manual encourages “providing copies of seized records” to the subject attorney “where such disclosure will not impede or obstruct the investigation” so that privilege issues can be determined. See U.S. Attorney’s Manual §9-13.420.

TAR requires a protocol and people to train the algorithm. Judge Maas’ proposal was that “the [TAR] process would continue as long as privileged data is identified by the TAR system” but did not explain how involved attorneys would be in teaching the program about privilege, instead suggesting that “[t]he Movants could assist [the] process by identifying documents that they consider privileged or otherwise protected.” Assuming Judge Maas meant that the TAR program would keep recalibrating until there were no new documents in a new iteration, he did not address how to ensure that all privileged documents were captured by the program.

Nor did Judge Maas’ proposal include any other methods of review for the electronic materials or a mechanism for the criminal defendant to challenge the tool’s privilege determinations. Given courts’ reticence to get involved in drafting TAR protocols, see Rio Tinto PLC, 306 F.R.D. at 129 (explaining that the TAR protocol “was the result of the parties’ agreement, not Court order”), they certainly shouldn’t be involved in designing a protocol in an arguably novel circumstance.

Court-ordered TAR may be unconstitutional. Fundamental due process rights dictate that the accused in a criminal proceeding is entitled to withhold legally-privileged information from the prosecution. Furthermore, the Constitution requires that the accused receive effective assistance of legal counsel. If courts were to require TAR for privilege review in criminal cases, it would risk ineffective assistance of counsel by restricting lawyers from dedicating adequate time and resources to protecting legal privilege. Indeed, when individual liberty is at stake, criminal defendants should not be subjected to the known or potential inaccuracies of a computer program and risk privileged material ending up in the prosecution’s hands.


Criminal defendants need to maintain control over their privilege review. Indeed, courts and prosecutors should not be dictating how criminal defendants review their documents for privilege. While TAR is an important discovery tool, it is not appropriate in all circumstances. Using technology can aid criminal defendants in their document reviews but they should not be at its mercy. At a minimum, a criminal defendant should not be required to accept the determinations of privilege reached by untested technology.

Ellen Murphy and Scott Morvillo are partners at Orrick Herrington & Sutcliffe in New York. Wendy Butler Curtis is chief innovation officer in the firm’s Washington, D.C. office, and Kelly Cullen is an associate in the Wheeling, W. Va. office. Summer associate Kristin Schwam assisted in the preparation of this article.

Reprinted with permission from the January 14, 2019 edition of the New York Law Journal © 2019 ALM Media Properties, LLC. All rights reserved. Further duplication without permission is prohibited, contact 877-257-3382 or [email protected]