Last Updated: October 30, 2023
DATE: October 30, 2023
Concerns Implicated by GenAI
Generative AI (“GenAI”) is a new technology that holds promise for making the conduct of discovery and the practice of law more efficient. However, this technology also raises a number of significant and material concerns, such as:
Accuracy: Fundamentally, the core large language models (“LLMs”) at the heart of GenAI-powered applications are designed to produce plausible-sounding text, not accurate information. Thus, there is the possibility of false, misleading, or entirely fabulated information appearing in the output of these systems. In legal use cases, failure to check GenAI results for accuracy may violate an attorney’s duty of competency.
Confidentiality: When using GenAI-powered applications, data must be sent to LLMs for processing. These LLMs are often developed and hosted by third-party vendors. In addition, these vendors may use this data to further train or finetune their language models, which can cause input from one customer to appear as output for another. The presence of third-party vendors and the possibility of input being used as training data raises important concerns about how use of GenAI affects an attorney's ethical duties of confidentiality, as well as broader privacy and security considerations.
Court Orders Addressing GenAI
These concerns have led some courts to implement checks around the use of GenAI in courtroom settings, primarily where GenAI is used in the preparation of court filings. Depending on the jurisdiction and subject to any forthcoming standing orders, attorneys and pro-se litigants who use GenAI may be required to:
identify the GenAI tool and manner in which it is used;
identify portions of text drafted with assistance of GenAI;
declare that all AI generated output will be verified for accuracy by a human being; and/or
declare that use of any AI tool did not result in disclosure of confidential or business information to any unauthorized party.
EverlawAI Assistant Enables Verification and Ensures Confidentiality
Everlaw is committed to helping our customers use technology responsibly. EverlawAI Assistant is a suite of features that leverage LLMs to deliver time-saving insights to legal teams as they review and synthesize documents and facts. Customers are in full control over if and when these features are used, and the output is indicated to the user as having come from a GenAI system (e.g., presented in a different font or denoted with an AI icon, etc.).
We know our customers host sensitive case materials with us, and can use Everlaw tools to create work product that may be incorporated into filings and representations made to courts or to other parties. For customers who wish to leverage Everlaw’s GenAI tools, but are concerned about compliance with court orders, this document addresses:
EverlawAI Assistant features that make it easy for customers to verify the accuracy of results before using them in meaningful ways; and
The technical design and contractual commitments we’ve implemented to mitigate privacy and confidentiality risks.
Mitigation of Inaccurate Results and Simplicity of Verification. Everlaw designs our GenAI features with the goals of (i) mitigating the occurrence of false, misleading, and entirely fabulated results, and (ii) making it simple for customers to apply human judgment to verify output.
To mitigate inaccuracies, Everlaw “grounds” the LLM in the text of documents, and information derived from it. For example, when asking a question of a document in AI Review Assistant, the LLM is instructed to base its answers on the text of the document. With the AI Writing Assistant, the LLM is instructed to produce output based on actual information about the documents it is being asked to analyze.
Besides “grounding” the LLM in document text and information, Everlaw designs the features such that the customer can easily trace the source of information the LLM is relying on to create its output. In the Review Window, users can easily hone in on the specific page ranges or portions of text underlying an insight generated by the LLM to verify its accuracy. Similarly, the Writing Assistant cites specific pieces of evidence. Everlaw turns these citations into direct links to the underlying documents so that customers can trace an assertion or fact to its evidentiary source.
While Everlaw enables quick and easy verification, the customer always remains responsible for leveraging these tools and design elements to actually verify generated information.
Technical and Contractual Data Protections to Ensure Confidentiality
Vendor’s contractual commitments: Everlaw uses OpenAI as our LLM vendor. This means customer data leaves the Everlaw environment to OpenAI servers for processing when the customer uses an EverlawAI Assistant feature. However, OpenAI is expressly prohibited from (1) training its models with this data or (2) retaining this data for any period after processing for any purpose (“zero data retention”). In other words, OpenAI accesses Everlaw customer data only for the amount of time necessary to perform the AI services requested and does not retain the data for any period of time after that for any reason. Further, in line with Everlaw’s confidentiality obligations to our customers, OpenAI has contractually agreed to take precautions to protect Everlaw customer data and not use or divulge Everlaw customer data other than as necessary to provide the contracted services to Everlaw.
Vendor technical controls: Everlaw’s (i) legal, (ii) security, and (iii) governance, risk, and compliance teams performed a thorough vendor assessment of OpenAI before Everlaw and OpenAI signed an enterprise level agreement. OpenAI is also certified SOC 2 compliant, meaning their controls have been determined by a third party to process client data in a secure manner.
Everlaw’s contractual commitments: Case materials used as inputs into Everlaw’s GenAI features, and the resulting output, are all customer “Case Materials” (as defined in the terms of service between Everlaw and the customer), and subject to the same security, privacy, and confidentiality commitments as the customer’s other case materials on the platform.1
Everlaw technical controls: When an Everlaw customer uses an Everlaw GenAI feature, each data request is sent securely to OpenAI individually, over an SSL encrypted service. When the results are returned from OpenAI, they are securely stored with the rest of the customer’s case materials, which are protected by strict access controls. Our systems are designed so that data is never shared between Everlaw customers.
This document will be updated from time to time to reflect requirements in new standing orders as we learn about them, or changes to Everlaw’s GenAI features.
The information provided on this page does not, and is not intended to, constitute legal advice; instead, all information, content, and materials available on this site are for general informational purposes only.
As of October 30, 2023, use of EverlawAI Assistant (which is in beta) is also subject to the Evaluation Agreement.