OpenAI adds governance tools

OpenAI has started shipping the plumbing for corporate AI use: a new open-weight Privacy Filter to scrub personal data, and new Codex partnerships to wire AI into enterprise software work. (openai.com) OpenAI said on April 22 that Privacy Filter detects and redacts personally identifiable information in text, including names and account details, before that text is sent onward. The company said the model can run locally, so unfiltered data does not have to leave a laptop or server first. (openai.com) The GitHub release says Privacy Filter is licensed under Apache 2.0, has 1.5 billion parameters with 50 million active at a time, and supports a 128,000-token context window. OpenAI said teams can fine-tune it and use it in training, indexing, logging, and review pipelines. (github.com) Privacy filtering is the step that strips out private details before a larger model reads a document, like blacking out names on a photocopy before sharing it. OpenAI said older tools often rely on fixed patterns such as phone-number formats, while Privacy Filter uses context to decide when a name refers to a private person and when it refers to a public figure or fictional character. (openai.com) A day earlier, on April 21, OpenAI said it was launching Codex Labs and expanding partnerships with Accenture, Capgemini, CGI, Cognizant, Infosys, PwC, and Tata Consultancy Services. The company said those firms will help customers move Codex from pilots into production across the software development lifecycle. (openai.com) OpenAI said more than 3 million developers were using Codex weekly in early April, and that figure had risen above 4 million by April 21. The same post named Virgin Atlantic, Ramp, Notion, Cisco, and Rakuten as companies already using Codex for tasks such as testing, code review, feature building, repository analysis, and incident response. (openai.com) Infosys said on April 22 that it will integrate OpenAI tools, including Codex, into its Topaz platform for clients working on software engineering, legacy modernization, and DevOps. TechCrunch reported the companies did not disclose financial terms. (techcrunch.com) That combination shifts the enterprise AI problem from “can the model write?” to “can the company control where data goes and how outputs enter existing systems.” OpenAI’s own descriptions of Privacy Filter and Codex Labs both focus on on-device redaction, workflow integration, and repeatable deployment inside existing engineering and compliance setups. (openai.com, openai.com) For media companies, that usually means connecting AI to content management, archives, rights systems, and legal review without exposing source material or personal data. OpenAI’s new releases do not solve those internal policies by themselves, but they do package two missing pieces: a local redaction layer and a services channel for fitting AI into production systems. (openai.com, openai.com) OpenAI is no longer just selling a chatbot or a coding assistant. This week’s releases show it selling the controls and consulting needed to get those tools inside large organizations. (openai.com, openai.com)