OpenAI Doubles Down on Zero Data Retention for Frontier Models
OpenAI has reaffirmed its commitment to Zero Data Retention (ZDR) for eligible API customers using its frontier models. This means that customer prompts and model responses are not retained after a request is processed. Furthermore, customer content is not accessible to OpenAI personnel for review, and enterprise customer data is not used to train OpenAI's models unless customers explicitly opt-in. This commitment is crucial for organizations that need to meet regulatory obligations, maintain customer trust, and protect their competitive advantage.
The expansion of ZDR is a significant step, especially as AI systems are tasked with increasingly complex and sensitive operations. While the specific criteria for ZDR eligibility for API customers are expected to be detailed in a forthcoming technical white paper in September, the announcement underscores OpenAI's dedication to robust data privacy.
Introducing Private Safety Processing for Enhanced Security
In conjunction with its ZDR commitment, OpenAI is previewing a new system called Private Safety Processing. This innovative system is designed to identify potential safety risks and patterns across related interactions without granting OpenAI personnel access to the underlying content. This is a critical development, as some risks may only become apparent when evaluating multiple interactions, rather than individual ones.
Private Safety Processing builds upon existing automated protections already in place for ZDR and other deployments. It extends these protections across related interactions, allowing automated systems to identify patterns without exposing customer content to OpenAI personnel. For ZDR deployments, customer content can remain on infrastructure controlled by the customer. OpenAI is also developing an option where content is stored on OpenAI infrastructure but encrypted with keys controlled by the customer, ensuring that OpenAI employees do not possess these keys and therefore cannot access the underlying content. When a risk is identified, OpenAI receives a narrowly defined signal indicating the type of activity, rather than the actual customer content, which is then used to determine if enforcement is necessary.
The Competitive Landscape of AI Privacy
The emphasis on enhanced privacy protections by OpenAI comes amidst a growing competition with other leading AI developers, notably Anthropic. This rivalry is particularly evident in the realm of enterprise customer data privacy. While OpenAI has moved to offer zero data retention, Anthropic has reportedly instituted a 30-day data retention policy for business customers using its Fable 5 and Mythos 5 models.
Both companies generally do not train their models on enterprise or API data by default, but consumer-facing services like ChatGPT and Claude may use chats for training unless users opt out. The push for stronger privacy measures reflects the increasing importance of data control for businesses adopting AI solutions, especially given the sensitive nature of information involved, such as financial records, proprietary code, and health data.
OpenAI's Broader Privacy and Security Framework
Beyond these new announcements, OpenAI maintains a comprehensive approach to data security and privacy for its business offerings. The company does not train its models on organizational data by default and offers enhanced data retention controls to assist with compliance. OpenAI's platforms are built with security and privacy in mind, undergoing testing by security experts and independent third-party auditors.
Key aspects of OpenAI's existing privacy and security framework include:
- Data Encryption: All data is encrypted at rest (AES-256) and in transit (TLS 1.2+) between customers and OpenAI, and between OpenAI and its service providers.
- Compliance: OpenAI supports compliance with major privacy laws such as GDPR, CCPA, HIPAA, and FERPA, offering a Data Processing Addendum and Business Associate Agreement.
- Access Controls: Strict access controls limit who can access data, and authorized OpenAI employees only access conversations for incident resolution, with explicit customer permission, or as legally required.
- Audit Logs: For enterprise accounts, workspace administrators can access an audit log of conversations and GPTs through the Enterprise Compliance API.
These measures collectively aim to provide businesses with greater confidence in deploying OpenAI's advanced AI capabilities while safeguarding their sensitive information.