Moderation Layer
ContractKen's proprietary 'Moderation Layer', built inside Word add-in provides a
secure, private and confidential way to use AI best-in-class AI models like GPT-4o, o1, o3, Claude Sonnet 3.5, Gemini 2.0 for contracts, within Word
The 'Moderation Layer' ensures that all text inside contract document is automatically anonymized and replaced with labels, in accordance with policies established by customer's IT departments, before the text is analyzed by any cloud service. This ensures that users can leverage ever more powerful models (available via API) without worrying about the privacy and confidentiality of their clients / organization being breached. The Infosec teams in the customer enterprises maintain control over data usage, with the assurance that all data interactions are securely logged and audited.
Read this blog for more details: https://www.contractken.com/post/moderation-layer
Audit Logs
ContractKen gives you granular audit logs on all data shared with AI providers. Understand who is sharing what data, with whom, with full transparency into the MSAs and Terms of Service governing that data.
Automatic Anonymization
ContractKen automatically redacts sensitive data (client names, dates, individual names, etc.), phrases, clause language, commercial information, or entire categories of data (SSNs, PHI etc.) before the data leaves your organization. Full control for Infosec teams. Frictionless for users.
WYSIWYG
ContractKen shows you the actual anonymized version of the document at the click of a button so that the user is comfortable about privacy. All of the anonymized versions are saved separately on our server to ensure a full audit trail - so that you are not only using LLMs safely but can also prove it, if needed later.
You define what is sensitive, private & confidential
Your Infosec admin teams have the ability to define what words, text, phrases fall into sensitive, private and confidential categories. Our algorithms automatically anonymize PII elements, financial information, PHIs, etc. You can add what constitutes proprietary information, custom clause language or even style and tone of language to be preserved.
Our plan around Data Privacy and Confidentiality
ContractKen is committed to making use of Generative AI 100% safe, private and secure for Contracts, from inside Microsoft Word. While we have implemented the industry leading ‘Moderation Layer’ solution, we are not sitting on our laurels. Our engineering team is working on testing the use of GANs (Generative Adversarial Networks) for preserving privacy.
Generative Adversarial Networks (GANs) can be used to generate synthetic data that preserves the statistical characteristics of the original data without compromising privacy. This approach can be particularly useful in preserving privacy and confidentiality when using Large Language Models (LLMs).
Here's a high-level roadmap:
1. Training GANs: First, train a GAN on a dataset of contract texts. The GAN consists of two parts: a generator and a discriminator. The generator's role is to create new synthetic contract text data, and the discriminator's role is to distinguish between real and synthetic data. The two networks are trained together, with the generator improving its ability to create convincing synthetic data, and the discriminator improving its ability to differentiate between real and synthetic data
.
2. Generation of Synthetic Contracts: Once the GAN is trained, you can use the generator to create synthetic contracts. These contracts will have similar characteristics and language as the original contracts but will not contain any real, sensitive information. The synthetic contracts can be generated to maintain the structural and statistical properties of the original contracts.
3. Analysis with LLMs: Then, you can use an LLM to review and analyze these synthetic contracts. Since these contracts do not contain any real sensitive data, the privacy and confidentiality of the original contracts are preserved
.
Generative Adversarial Networks (GANs) can be used to generate synthetic data that preserves the statistical characteristics of the original data without compromising privacy.
