IronCore Labs launches VectorLens to scan hidden PII in AI embeddings

IronCore Labs has released VectorLens, a local command-line tool that scans vector embeddings directly to find personally identifiable information and other sensitive data inside AI systems. The launch targets privacy, security, and governance teams trying to uncover shadow data in RAG pipelines and vector databases.

Why it matters: - AI embeddings can carry approximate copies of source data, turning vector databases into hidden stores of customer records, support tickets, HR files, financial details, and health information. - Security, privacy, and GRC teams need a way to find sensitive data already embedded in AI workflows before it becomes an exposure or compliance issue. - VectorLens is designed to help organizations identify shadow data, audit unmanaged vectors, and decide when monitoring, governance, or encryption is needed.

What happened: - IronCore Labs announced the availability of VectorLens on June 30, 2026. - VectorLens is a private command-line tool that scans AI vector embeddings locally to discover and classify personally identifiable information and other sensitive data hidden in vector databases. - The tool is available now and free to try.

The details: - VectorLens scans exported vectors directly instead of requiring access to source text or labels created before embeddings were generated. - The tool can inspect orphaned, imported, unlabeled, or unmanaged embeddings even when the original source data is unavailable. - VectorLens runs locally on macOS and Linux as a self-contained binary. - Users export embeddings from a vector database, run a scan, and receive a report showing which vectors contain sensitive data. - Output options include inline results, machine-readable JSON, and a PDF report for security, privacy, compliance, and executive teams. - VectorLens works with any vector store that can export to JSONL or Parquet, including Pinecone, Weaviate, Milvus, Qdrant, Elastic, and Postgres/pgvector. - Launch support covers common embedding models including all-minilm-l6-v2, bge-m3, gtr-t5-base, text-embedding-ada-002, and text-embedding-3-large. - The tool can classify names, email addresses, phone numbers, physical addresses, dates of birth, credit card numbers, social security numbers, numeric identifiers, and other sensitive content. - IronCore Labs said it only performs lightweight network calls for license validation and high-level usage metrics. - IronCore Labs said it never sees the vectors or private data being scanned because the tool runs in the customer’s own environment. - Teams can learn more, request a license key, and start scanning vectors on the IronCore Labs website.

Between the lines: - The launch reflects a broader shift in AI security from protecting source systems to finding sensitive data after it has been transformed into embeddings. - The focus on local scanning suggests demand for tools that can work without moving private data into another vendor environment. - VectorLens also positions vector encryption as a possible next step for organizations that find exposure in their AI pipelines.

What's next: - IronCore Labs plans additional model support beyond the launch list. - Organizations using RAG systems and vector databases can use VectorLens to baseline exposure, watch for regressions, and audit unmanaged stores. - Teams that find sensitive data in embeddings may use the results to decide whether governance changes or vector encryption are needed.