Mirror Security and Hoplo, the company behind Teriyaki, have announced a strategic partnership to bring encrypted AI to every stage of how Teriyaki builds and runs software for companies.
Teriyaki allows people within a company to describe a process in their own words and receive software that carries it out. Each application is verified by a separate validator, approved by a person and connected to the systems the company already uses. To make this possible, Teriyaki works with a company’s most sensitive material, including its procedures, contracts, price lists, customer records and internal know-how.
Under the partnership, Mirror’s fully homomorphic encryption (FHE) engine, which allows AI to compute on data without ever decrypting it, will be integrated into the Teriyaki platform, keeping this material encrypted end to end. The integration works across three layers.
During the build stage, the description of a process, its specifications and the generated code travel to AI models in encrypted form through Mirror Encrypted Token Factory, and the code stays encrypted at rest, in transit and during the inference that produces it. When an application built with Teriyaki calls an AI model or works on company records, inference runs on encrypted data through Mirror VectaX, ensuring that personal, financial and health data are never exposed in clear text to the model. The knowledge Teriyaki gathers about how a company works, including documents, processes and the embeddings used to search them, is stored and queried in encrypted form. Since retrieval happens on encrypted vectors, this company brain remains useful to the organisation while staying unreadable to anyone else.
The integration is designed to add no friction for developers or end users. Mirror ETF connects through an OpenAI-compatible API without requiring code changes, and every call generates a signed run receipt that provides a verifiable record of what the AI did with company data.
Beyond encryption, Mirror AgentIQ governs the agents and tool connections Teriyaki uses, including Model Context Protocol (MCP) connections, applying identity controls and runtime policies in under 25 milliseconds and recording each decision as a signed verdict. Before any application is published, it must clear two checks: Teriyaki’s validator confirms that the software works, while Mirror DiscoveR tests it against more than 130 adversarial attack techniques. Encryption keys remain with the customer throughout.
The joint offering is aimed at organisations that want to automate their knowledge work with AI without handing their data to an AI provider.
“Companies want AI to do real work, but the first question they ask us is always the same: where does our data go? With Mirror, the answer is simple. Their code, their documents, and their know-how stay encrypted from the moment Teriyaki starts building to every time the software runs. The company brain belongs to the company, and only the company can read it,” said Alessio Bagnaresi, CEO and Co-Founder of Hoplo, the company behind Teriyaki.
“Teriyaki turns a company’s knowledge into working software, and that knowledge is exactly what attackers and careless pipelines put at risk. Most AI security protects data until the model needs it. Mirror lets the model work on data it can never read, so Teriyaki customers can hand AI their most sensitive work and still keep the keys,” said Pankaj Thapa, Co-Founder and CEO of Mirror Security.
The partnership also supports companies in meeting their obligations under the GDPR and the EU AI Act. Signed run receipts and policy verdicts give teams an auditable record of every AI action, ensuring privacy and governance are built in from the start.
About The Author
