Stay updated with the latest industry-leading AI coverage by subscribing to our daily and weekly newsletters. Learn More
Enterprises are witnessing a surge in the unauthorized use of AI tools by their employees, a phenomenon known as shadow AI. Studies show that up to 96% of AI-related work done by employees is through non-corporate accounts, posing a significant risk of leaking sensitive corporate data.
Cyberhaven, a security platform, aims to address this issue by tracking data lineage across different users and endpoints. Through its large lineage models (LLiMs), Cyberhaven has introduced Linea AI, a next-generation platform designed to combat shadow AI and predict potentially harmful incidents.
Nishant Doshi, Chief Product and Development Officer at Cyberhaven, explained, “Lineage allows you to understand the origin of data, its access history, and its movement across various endpoints and users.”
90% Reduction in Incidents Requiring Manual Review
Recent analysis by Cyberhaven revealed a 485% increase in AI usage among employees between March 2023 and March 2024. Alarmingly, a large percentage of sensitive data such as legal documents, source code, and HR records shared with AI is directed towards non-corporate accounts.
Linea AI leverages LLiMs trained on real enterprise data flows to prevent unauthorized data sharing and safeguard company information. The platform incorporates computer vision and multi-modal AI to analyze various data formats, enabling autonomous assessment of policy violations and incident severity to alleviate security operations center alert fatigue.
As a result, customers have reported a 90% reduction in incidents requiring manual review and an 80% decrease in mean time to respond to security incidents. Cyberhaven’s tools have uncovered over 50 critical risks per month that traditional security measures failed to detect.
Prabhath Karanth, CSO and CIO of Greenlight, a family financial app, praised Cyberhaven for providing visibility into data movements across the organization, surpassing traditional data loss prevention (DLP) and insider risk management tools.
Unlike conventional approaches that rely on pattern matching, Cyberhaven emphasizes content and context inspection by examining data and providing context through lineage traces.
Protecting Enterprises’ Valuable Data with AI
Cyberhaven’s cutting-edge AI models and transformer neural network architecture power its platform, enhancing data analysis accuracy. Employing a multi-stage retrieval-augmented generation (RAG) engine, Cyberhaven fine-tunes its LLiM to identify an enterprise’s most critical data elements.
The platform introduces intelligent screenshot analysis to address a persistent blind spot in data security. Aaron Arkeen, a senior security engineer at DailyPay, highlighted the challenge of detecting and preventing the unauthorized sharing of sensitive information such as engineering designs and research data.
Enhancing User Monitoring
Cyberhaven introduces an autonomous AI feature, Let Linea Decide, to assist security teams in assessing incident severity by analyzing data and user logs. This feature offers insights into screenshots, PDFs, and source code, leveraging data lineage to determine the necessity of human intervention.
Doshi explained, “We aim to predict the next action based on historical knowledge, distinguishing between benign and anomalous events through data comprehension.” The platform aids in identifying suspicious user activities and reducing manual review efforts.
By detecting incidents like data exfiltration to personal cloud accounts and unauthorized file transfers, Cyberhaven empowers organizations to prevent data breaches proactively. Arkeen noted that DailyPay saw a 65% reduction in mean time to respond due to Linea’s concise AI summaries and efficient risk identification.
Despite evaluating multiple DLP providers, DailyPay chose Cyberhaven for its unique data lineage approach, which offers unparalleled visibility into data movements and risk mitigation.
Arkeen concluded, “Linea AI consistently identifies nuanced risks that traditional systems overlook, saving us time on escalation and prevention efforts.”
