OpenAI is previewing Private Safety Processing for its frontier large language models (LLMs) in a move to give enterprise customers more control over how their data is handled while allowing it to monitor for potentially risky activity.
Private Safety Processing allows the AI model developer to support Zero Data Retention (ZDR), whereby a service provider processes a customer’s data in real-time and discards it immediately rather than storing it for a period of time. The company stated:
“OpenAI does not retain their prompts or model responses after a request is processed. Customer content is not available to OpenAI personnel for review, and enterprise customer data is not used to train our models unless customers explicitly opt-in.”
The move comes as OpenAI pushes deeper into the enterprise market, where data privacy and control are key requirements for organizations deploying increasingly capable models in sensitive workflows. It also comes amid reports that OpenAI is preparing for a potential initial public offering (IPO), putting greater focus on its ability to turn enterprise demand into commercially sustainable growth.
Balancing AI Safety With Data Privacy
The announcement made a not-so-subtle dig at competitor Anthropic, noting that “[s]ome recent frontier-model deployments have required customers to allow their AI provider to retain sensitive content for safety monitoring.” While model providers may need access to more context to identify misuse, such policies can restrict some enterprises from adopting AI services where retaining sensitive customer data would conflict with their security policies, regulatory requirements, or commitments to protecting customer information.
“Enterprise AI adoption depends solely on customer control of data, with no direct or derivative use beyond the chosen service,” said Sunil Agrawal, Chief Information Security Officer at Glean.
As Ashish Nagar, Co-Founder and CEO at Level AI, told CX Today:
“Enterprises will need control of how their data is being used, where how is it being stored, and how is it being used for model training to create a continuous learning loop.”
OpenAI said it is responding to those concerns. “We are sharing this preview now because we’ve heard our customers loud and clear that they need predictability about how their content will be protected as AI systems become more capable.”
“The organizations we work with handle some of the most sensitive information in their sectors, including financial records, health data, confidential business plans, and proprietary research. Protecting that information is essential to meeting regulatory obligations, maintaining customer trust, and preserving their competitive advantage.”
The issue is particularly relevant to customer experience teams, which increasingly use AI to process data from customer interactions. Call transcripts, complaints, account information and other customer records can contain sensitive personal and commercial information.
The risk extends beyond formally approved AI deployments. The need for tighter data controls is increasing as enterprise employees turn to AI chatbots and agents for everyday work, potentially moving customer information outside the systems and safeguards controlled by their employers. Customer data can end up in personal or unsanctioned AI tools when employees lack convenient enterprise-approved alternatives, creating a gap between where organizations believe data is being handled and where it actually goes.
OpenAI’s Private Safety Processing system offering addresses part of this concern by giving eligible API customers stronger assurances over how prompts and responses are retained when employees use its models.
Research by data privacy company Incogni indicates that OpenAI carries lower risk than some of its competitors, placing ChatGPT among the most privacy-friendly platforms assessed. The Gen AI and LLM Data Privacy Ranking 2026 gave ChatGPT the second-lowest overall privacy-risk score and found OpenAI offers the clearest information about it handles user data, although Incogni noted that opting out of training does not remove information that has already been incorporated into a model.
Monitoring Longer-Running AI Agents
OpenAI said the Private Safety Processing system, currently being tested with certain early customers, is designed to preserve ZDR commitments while allowing automated systems to detect risks from content that may only become apparent across multiple interactions or during longer-running agentic tasks.
