OpenAI adds privacy layer
- OpenAI released Privacy Filter, an open-source on-device model that redacts personal data from enterprise datasets. - The company is also reported to be in talks to commit up to $1.5 billion to a private-equity joint venture for enterprise AI deployment. - Those moves pair a privacy product with financing and distribution efforts, indicating enterprise AI sales require both data controls and deployment scale (venturebeat.com, reuters.com).
OpenAI has released a new model that strips personal details out of company data before that data leaves a laptop or server. (openai.com) The tool is called Privacy Filter, and OpenAI published it on April 22, 2026 as open source under the Apache 2.0 license. OpenAI’s GitHub repository says it is built for “high-throughput data sanitization workflows” that can run on-premises on central processing units or graphics processing units. (openai.com, github.com) In plain terms, the model scans text for names, account numbers, contact details and other personally identifiable information, then marks or redacts those spans before the text is used elsewhere. OpenAI says the released model has 1.5 billion parameters, 50 million active parameters, and a 128,000-token context window, which lets it process long documents without splitting them into many smaller chunks. (openai.com, huggingface.co) That matters for companies that want to use large language models on support logs, legal files, health records or internal documents without first sending raw personal data to an outside provider. OpenAI said the model is small enough to run locally so unfiltered data can remain on device rather than being shipped to a server for de-identification. (openai.com, venturebeat.com) The privacy release landed one day before Reuters reported that OpenAI is in talks to commit up to $1.5 billion to a new private-equity joint venture aimed at enterprise artificial intelligence deployment. Reuters, citing the Financial Times, said OpenAI would initially invest $500 million of equity into the venture, known internally as DeployCo, and could raise that commitment to $1.5 billion. (reuters.com) The same Reuters report said the joint venture is expected to be valued at $10 billion in a funding round targeted to close in early May 2026. It named TPG, Bain Capital, Advent, Brookfield and Goanna Capital as firms in talks to participate. (reuters.com) Taken together, the two moves point at the same enterprise sales problem: companies need both a way to clean sensitive data and a way to finance and install large systems. VentureBeat described Privacy Filter as a local-first privacy layer for enterprise workflows, while Reuters described DeployCo as a vehicle for broader rollout of enterprise artificial intelligence infrastructure. (venturebeat.com, reuters.com) OpenAI said it already uses a fine-tuned version of Privacy Filter internally, and the public release is meant for jobs like training-data cleanup, logging systems and review pipelines. The company’s GitHub page says users can tune precision and recall settings, which means choosing whether the model should catch more possible private data or avoid over-redacting text that is not sensitive. (openai.com, github.com) For enterprise buyers, the pitch is no longer just access to a model. It is a package of privacy controls, local processing and capital to get systems deployed at scale. (openai.com, reuters.com)
