OpenAI pledges zero data retention to lure Anthropic's 30-day hold customers

OpenAI is betting that a zero-data-retention promise can pry enterprise clients away from Anthropic, whose 30-day data-holding policy has drawn fire from Silicon Valley and the White House.

OpenAI's pledge not to retain business customer data directly targets Anthropic's 30-day retention policy, a bid to win enterprise clients in regulated industries as the two labs escalate their rivalry for AI dominance.

"We've heard very loud and clear from businesses that this is important. They often have their own commitments that they have made to their customers if you think about somebody who is serving other enterprises, who are trusting them with incredibly sensitive data," Aleah Houze, OpenAI's head of product policy, said.

The maker of ChatGPT previewed "Private Safety Processing" on Aug. 19, technology that detects safety risks across multiple interactions without giving OpenAI personnel access to the underlying content. Customer data stays on infrastructure the customer controls, or on OpenAI storage encrypted with keys the customer owns. When a risk such as a cyberattack or biological-weapon activity is flagged, OpenAI receives only a high-level signal detailing the severity and category of the threat.

The move comes as OpenAI's second-quarter revenue growth lagged Anthropic's, disappointing some shareholders, The Wall Street Journal reported. Microsoft, OpenAI's largest backer, and Palantir, whose chief executive Alex Karp has criticized data retention, are among the companies watching how the privacy battle reshapes enterprise AI spending.

The 30-Day Policy That Opened a Door

Anthropic's popularity with businesses that pay for its coding tools and advanced Claude models has vaulted the company ahead in the AI race, but its practice of keeping customer data for 30 days to safeguard its newest models has fueled a backlash. White House AI adviser and venture capitalist David Sacks said in June that retaining data was "creating a new level of AI haves and have-nots," fueling outrage among software developers. Microsoft Chief Executive Satya Nadella and Palantir's Karp have argued that businesses, particularly in healthcare and finance, need certainty that AI companies are not keeping their data.

OpenAI's answer is a system that extends its existing Zero Data Retention guarantee, which already promised eligible API customers that prompts and model responses would not be retained after a request was processed. The new Private Safety Processing, now being tested with early customers including Microsoft, Databricks, Glean and Abridge, lets automated systems identify patterns across related interactions — catching a user who probes a company's software weaknesses in one conversation and asks about remote-access tools in another, for example — without exposing the content to OpenAI staff.

Customers can investigate alerts using their own systems and may share information with OpenAI if they want to appeal an enforcement decision or support a probe into verified abuse. OpenAI keeps only the high-level signal, and personnel do not receive access to customer content even when it is flagged. The company plans to roll out the system and publish a technical white paper in September.

The Open-Weight Debate Looms

The privacy fight is tangled in a separate dispute over open-weight models, which let users download the numerical parameters that determine behavior. Anthropic has warned these tools pose safety risks because they often lack the safeguards of the most powerful closed models. OpenAI, which signed an industry letter favoring open-weight models and releases some of its own, has held talks with administration officials about potential restrictions on Chinese open-weight providers accused of using U.S. advances through a process called distillation.

For investors, the stakes are concrete. OpenAI remains private, so the direct equity exposure runs through Microsoft, which has committed billions to the lab and embeds its models across Azure. A win in enterprise data privacy could slow Anthropic's momentum in the coding and agentic-tool market, where its Claude models have gained ground. Sunil Agrawal, chief information security officer at Glean, said the no-training commitment and ZDR give his company "confidence to build with OpenAI," a signal that data control is becoming a decisive factor in enterprise AI procurement.

This article is for informational purposes only and does not constitute investment advice.