Metomic, a data security platform, has introduced two new AI-powered solutions, Semantic Asset Classification and Data Cleanser, designed to enable enterprises to safely integrate artificial intelligence tools while safeguarding sensitive information. These solutions aim to address critical security vulnerabilities that arise when organizations incorporate AI agents and large language models into their workflows.
The increasing adoption of AI tools such as Gemini Gems, Dust, and Microsoft Copilot by enterprises has highlighted significant risks of inadvertently exposing sensitive data. Metomic’s research indicates that AI agents can easily extract confidential information, including employee emails, financial documents, and intellectual property, when provided with unredacted datasets. Ben van Enckevort of Metomic stated, “The magic trick we demonstrated shows the core problem every company faces with AI deployment. When you ask an AI tool to ‘give me all the emails referenced in this dataset,’ it will comply without hesitation – exposing sensitive information that should never be accessible.”
Semantic Asset Classification, the first solution, automatically identifies and labels entire documents based on their content. It supports categories such as Board documents, Financial data, HR data, and Intellectual property. The technology combines keyword detection with AI model validation, utilizing multiple frontier models to confirm classifications with high confidence levels. Dane Stevens, who led the development, explained, “Rather than looking for individual detections within documents, we examine the document as a whole and attach appropriate labels. This gives organizations unprecedented visibility into what types of documents they have, where they’re shared, and who can access them.”
The Data Cleanser addresses the need to redact sensitive information before data is fed into AI tools. This solution processes data from various sources, including Google Drive and Slack channels, automatically removing emails, phone numbers, and other personally identifiable information. In demonstrations, the tool successfully sanitized over 7,000 messages from Metomic’s internal development support channel while preserving contextual information necessary for AI training. Sandro Dolidze commented, “We can now take sensitive data sources like internal Slack channels and safely prepare them for AI tools. The Data Cleanser redacts all sensitive information while maintaining the utility of the data for AI training and analysis.”
Both solutions are currently in beta with select enterprise customers, with general availability projected for July 2025. Future roadmap items for the platform include custom classification labels, batch processing capabilities, and advanced truncation strategies. The platform already supports integration with Google Workspace, Microsoft 365, Zendesk, and various cloud storage platforms. Early access can be requested via Metomic’s website.
