Lest anyone underestimate the complexity of privilege searching in e-discovery, consider the recent case in which none other than search giant Google got tripped up. Although the company’s privilege searches screened out an email labeled “Attorney Work Product,” the searches failed to catch nine drafts of the same email, autosaved by the author’s email program.

Worse yet, the email was a true “smoking gun.” In fact, the email was so potentially inculpatory that the judge remarked that it and the Magna Carta (common law) would be all that the opponent’s counsel would need to win its case. (It is interesting to note that this was not the only incriminating document, and this document is even more damaging to Google when paired with a 2005 email written by Andy Rubin, Google vice president in charge of Android.)

In this post, we’ll give you some background on the case and then discuss the lessons it teaches about privilege searching in e-discovery.

A Dispute Over Java

Since August 2010, Google has been locked in litigation with Oracle America over whether its Android operating system violates Oracle’s Java patents and copyrights. Within the month before Oracle filed the lawsuit, lawyers for both companies met to discuss the alleged infringement. Ten days later, Google General Counsel Kent Walker convened an internal staff meeting to discuss the matter further. One of the attendees was Google software engineer Tim Lindholm.

A week later, as a follow-up to that meeting and a prelude to another, Lindholm wrote an email addressed to Google senior counsel Ben Lee and Andy Rubin. The email was labeled “Attorney Work Product” and “Google Confidential.” In part, it said:

What we’ve actually been asked to do (by Larry and Sergei) is to investigate what technical alternatives exist to Java for Android and Chrome. We’ve been over a bunch of these, and think they all suck. We conclude that we need to negotiate a license for Java under the terms we need.

During the four minutes it took Lindholm to write this email, his computer auto-saved it nine times. Each time, the “to” and “cc” fields were still blank and the email was not labeled with the footer. Only the tenth and final version identified the recipients and included the labels for “work product” and “confidential.”

During discovery after the lawsuit commenced, Google produced the first eight drafts of the email to Oracle. It held back the ninth draft and the final version, listing both on its privilege log. Google later told the court that its electronic scanning mechanisms “did not catch those drafts before production” because the drafts did not contain the confidentiality or privilege headings and did not list any addressees.

In court documents, Google said, “Due to the volume and speed of production in this case, Google has been forced to rely on electronic screening mechanisms, which identify potentially privileged documents based in part on sender and recipient information, as well as privilege-related keywords.”  It appears that these drafts slipped through because there was no “eyes on” review of the documents unless they hit on privilege terms.

The Emails Come to Light

On July 21, 2011, in two separate hearings in the case – one a telephonic hearing before U.S. Magistrate Judge Donna M. Ryu to compel Lindholm’s deposition and the other a Daubert hearing before U.S. District Judge William Alsup – Oracle referenced one of the email drafts. In the Daubert hearing, Oracle read part of it into the record. Google’s attorneys responded by addressing the substance of the email, but they did not object to it as privileged or confidential.

Judge Alsup readily picked up on the email’s significance. At one point during the hearing, he said to Google, “You are going to be on the losing end of this document” with “profound implications for a permanent injunction.” At another point, he said that a good trial lawyer could use the simple combination of the Lindholm email and the Magna Carta to win Oracle’s case and get an injunction. And to add insult to injury, all this took place with several news reporters in attendance, without objection.

Judge William Alsup

After those hearings, later the same day, Google sent notice to Oracle that the draft email constituted “protected material” under a protective order/clawback agreement and asked that it no longer refer to it in public. The next day, asserting that the draft was “unintentionally produced privileged material,” Google clawed it back. Soon after, Google clawed back all draft versions of the email.

Google’s next move was to ask Judge Alsup to redact the transcript of the Daubert hearing to remove all references to the Lindholm document. The judge denied the request, ruling that the email was protected neither by the attorney-client privilege nor the work-product doctrine.

Meanwhile, Oracle filed a motion with Judge Ryu to compel Google to re-produce the emails it clawed back. Judge Ryu granted Oracle’s motion, finding that the email was neither privileged nor otherwise protected. Google appealed that finding to Judge Alsup, who entered an order on Oct. 20, 2011, affirming the magistrate judge. “The Lindholm email and drafts will not be treated as protected by attorney-client privilege or work-product immunity,” he concluded.

Google now says it will appeal Judge Alsup’s ruling.

What This Means for Privilege Search

Although much can be said about whether or not these documents were privileged, whether the privilege was waived, and whether the actions of the attorneys in the hearings were well thought out, we will focus on the implications for search.

Producing Autosave Documents. One big question this case raises is whether producing parties should ever include “autosave” drafts of documents. Many software packages delete the autosave information when the user saves the document or sends the email. The answer will depend on the circumstances of the case, but it is a question counsel should always consider.

Clawback Agreements under Rule 502. The only sure way to protect privileged documents is to review them. While it was the goal of Rule 502 of the Federal Rules of Evidence to obviate the need to review every document for privilege, this case once again shows that “you can’t un-ring a bell.” Once the other side sees a damaging privileged document, it’s almost impossible to control the damage.

Keyword Searching and Highlighting. Keyword searching would have flagged this document as Potentially Privileged only if the search terms had included “negotiat*” or “licens*”. Many firms don’t like to use such broad terms because they flag too many documents as Potentially Privileged. At Catalyst, we have developed a list of several hundred words commonly used in legal documents, and many clients like to use them in their “Potentially Privileged” searches. However, doing so will identify a large number of “false hit” documents that are not privileged. One approach we recommend is to use term highlighting so that all such terms are easy to see as the document is reviewed. Even that wouldn’t have helped in this case because it appears that only documents that hit on privilege search terms were reviewed for privilege.

Near Dupe Analysis. We generally recommend using email threading and near-dupe analysis software. While the principal benefit is speeding the review, a second important benefit is to prevent inconsistent coding. In this case, when the attorneys were reviewing the final version of the emails for privilege, they would also have seen the autosaved drafts and could “tag all” as privileged. Further, we always recommend an “inconsistent coding” analysis, part of which is to identify documents in the same set of near dupes/email threads that are inconsistently coded for privilege. This would have caught the autosaved drafts before production in this case, where non-hits weren’t reviewed.

(It is interesting to note that in the most famous case of inadvertent production of a privileged, “smoking gun” document, Mt. Hawley Insurance Co. v. Felman Production, it appears that the damaging document was a dupe of a document withheld and on the privilege log. This underscores the importance of doing an inconsistent coding analysis! On this point, see also our earlier post: Bad Facts Make Bad Law: ‘Mt. Hawley’ A Step Backward for Rule 502(b)).

Other inconsistent coding checks. In addition to duplicate, near-dupe and attachment family inconsistent coding checks, other  methods we recommend to look for similar documents are “More Like This” and Key Document Clustering. In this case, this was such an important document that the attorneys should have used analytics to find similar documents as a QC measure.

Predictive Coding. Would predictive coding have helped? It’s hard to say. If the final versions of the emails had been used as seed documents, or tagged as privilege in the set of documents in the privilege sample set, there’s a good chance that the autosaved versions would have been highly ranked for privilege. On the other hand, if they were not in the set chosen to be sampled and not among the seed documents, then they might have slipped through.

The Honor System. It is also worth noting that in most cases the claims of privilege for documents listed on the privilege log are very seldom challenged.  The attorneys producing the documents are essentially on the honor system.  Had the “autosave” emails not slipped through, Oracle would not have had any way to challenge the claim of privilege for the withheld emails.  The claim of privilege in this case was certainly arguable, but it is the practice of many parties to withhold whole families of documents even when only one document is privileged and otherwise to err on the side of claiming privilege.

The Bottom Line

Recently, IDG News Service reporter James Niccolai wrote a thoughtful examination of the Google-Oracle privilege fiasco, How Google Was Tripped Up By a Bad Search. As he noted, the irony in the case is that “the e-mail might never have seen the light of day if the search tools used to identify documents covered by attorney-client privilege had done their job.” In any other case, that might not be considered ironic. But this, after all, is Google.

Whether these draft emails slipped through as a result of human or technological error, Google attributed the error to “electronic scanning tools.” Google has not provided further details and Judge Alsup’s opinion contained no further explanation.

Even so, as we’ve reviewed in this post, the case is not without lessons for e-discovery professionals. No human or technology can offer 100 percent protection against inadvertent production of privileged documents. However, some of the steps we’ve outlined above will certainly help minimize the risk.