OpenAI has pledged to avoid retaining any data from enterprise clients using its artificial intelligence models, while simultaneously enhancing its product safety capabilities, in a strategic move designed to lure customers away from rival Anthropic. Many businesses have expressed frustration over how the developer of the Claude model handles their operational data. On Wednesday, the creator of ChatGPT revealed that it has rolled out a preview of new technology to select clients, which is capable of identifying behavioral patterns and security threats across multiple rounds of model interactions.

Historically, AI safety systems could only monitor individual interactions in isolation, making it challenging to guarantee compliant tool usage without storing some client data; the new approach is intended to overcome this limitation. This commitment stands in stark contrast to Anthropic's current policy, which retains customer data for 30 days to ensure the secure operation of its latest models. That practice has drawn sharp criticism from several prominent tech leaders, including White House AI advisor and venture capitalist David Sacks, Microsoft CEO Satya Nadella, and Alex Karp of Palantir Technologies. They have argued that enterprises need absolute certainty that AI vendors will not preserve their data, a requirement that is especially critical in highly regulated sectors such as healthcare and finance.

Aliya Hauser, OpenAI's head of product policy, stated: "The message from enterprise clients is unmistakable, and this issue is of paramount importance to them. Many of these companies are themselves accountable to their own customers for data stewardship; these service providers that cater to other businesses hold vast amounts of highly sensitive information."

This initiative reflects OpenAI's ambition to close the gap with Anthropic in the AI race while also addressing criticism that its products lack sufficient safeguards against cyberattacks or various harms stemming from model malfunction. Reports indicate that OpenAI recently informed investors that its second-quarter revenue growth lagged behind Anthropic's, which disappointed some shareholders.

OpenAI has confirmed that customer data will reside within infrastructure fully controlled by the clients themselves. The company is also developing an additional system where data may be stored by OpenAI, but the encryption keys remain in the hands of the customer. This arrangement ensures that OpenAI employees cannot access user prompts, model responses, or other sensitive information. When the system detects high-risk activities from model usage, such as cyberattacks or bioweapon-related threats, it transmits a signal to OpenAI to help determine whether intervention is necessary. Clients are permitted to appeal any resulting decisions and may voluntarily supply additional information to OpenAI as needed. The company only retains this highly summarized signal record, which merely notes the risk level and risk category.