Enigmata raises $6.5M to run AI on encrypted data

Every small business that has held off on AI because of customer data has run the same math: the useful data is the sensitive data. Enigmata emerged from stealth on September 10 with $6.5 million in seed funding and a claim aimed squarely at that math — a cryptographic layer that lets models train, search, and analyze records while the records stay encrypted. The claim is worth understanding. The product is not yet worth buying.

What actually happened

Per the company's announcement, the Nashville-based startup raised a seed round led by Blockchange Ventures to commercialize Enigmata Cipher, described as patent-pending technology that converts records, documents and datasets into an encrypted form that AI, analytics and search tools can operate on using existing enterprise hardware — no specialized silicon.

Two claims stand out. First, in Enigmata's internal benchmarks, models trained on Cipher-protected data matched the accuracy of models trained on plaintext while training 8–10% faster. Second, Cipher supports removing an individual record without retraining the model from scratch.

Both numbers are vendor-reported, from internal testing, with no third-party evaluation published and no technical paper we could find. Cipher is currently limited to selected enterprise design partners. Target customers named are banks, insurers, health systems, life sciences firms and publishers. Treat the benchmarks as a direction the company is pointing, not a result you can plan around.

Why encrypted AI matters for your business

Ignore the round. Look at the second claim, because it names a problem most teams have not priced: deletion. When a customer invokes a right-to-delete under GDPR or CCPA, you can drop the row from Postgres in a second. You cannot drop it from a fine-tuned model's weights, or from the embeddings sitting in your vector store, or from the six months of prompt logs your provider retained. Most small operations discover this the first time a deletion request arrives after they shipped a RAG feature.

You do not need a seed-stage cryptography vendor to handle it. You need a data map. Know which customer fields ever reach a model, which embeddings are derived from them, and which store holds those embeddings with a key back to the customer ID. Then deletion is a job that clears a row, deletes the matching vectors, and rotates the cached context — not a research problem. Do it now, while your index is small.

The second thing to fix today costs nothing: turn off training on your data at the provider level and confirm the retention window in writing. That is where most exposure actually lives, not in the weights.

Sources: Enigmata emerges from stealth with $6.5 million (PR Newswire), The SaaS News: Enigmata raises $6.5M seed.