WHY THIS MATTERS
OpenAI announced on August 19, 2026 the expansion of Zero Data Retention, or ZDR, for its frontier models, a configuration under which the company commits to not keeping a copy of prompts and responses once a call finishes processing. The feature covers endpoints including chat completions, the Responses API, image generation and edits, embeddings, audio, moderation and the realtime API, but it only turns on when enterprise customers request it through OpenAI's sales team, it is not on by default. The company's own documentation notes that logs can still be kept when required by law or deemed necessary to protect its services, even under ZDR.In this article
What Changes With Zero Data Retention
On August 19, 2026, OpenAI formalized the expansion of its Zero Data Retention setting, known as ZDR, to its frontier models. In practice, ZDR means the prompt sent and the response generated are processed in memory and discarded as soon as the API call completes, without entering the standard 30 day retention window used for abuse monitoring, without feeding training corpora, and without being used for fine tuning or aggregated service quality analysis.
Which Endpoints Are Covered
The setting now applies to a broad set of platform endpoints: chat completions, the Responses API, image generation and edits, embeddings, audio processing, content moderation, the legacy completions endpoint, and also the realtime API used for voice and streaming applications. That covers practically the entire catalog of features a customer service or marketing automation company would tend to use in production.
Not a Switch Any Account Can Flip on Its Own
The detail that often gets missed is that ZDR is not a feature that simply appears ready to use on any account's dashboard. Enterprise customers need to request activation directly through OpenAI's sales team, endpoint by eligible endpoint, and go through an approval process before the no retention guarantee starts applying to the account's real traffic. Companies that mistakenly assume the setting is already active risk operating under the standard retention policy without realizing it.
The Caveat OpenAI Itself Highlights
OpenAI's official documentation is direct about the limits of the promise: even with ZDR turned on, logs can still be kept when the law requires longer retention, or when the company reasonably judges it necessary to keep data to protect the integrity of its services. What the feature actually removes is the standard 30 day window routinely used for abuse monitoring, not an absolute, unconditional guarantee of zero storage under every scenario.
Why the Timing Matters
The timing of the announcement is not incidental: AI oversight has been intensifying across several regulatory fronts at once, from the European AI Act's enforcement starting in early August to Brazil's ANPD now demanding documented proof of compliance from AI systems that process personal data. Companies running AI agents on top of customer data increasingly need to demonstrate explicit control over where that information is stored and for how long, and a feature like ZDR becomes concrete evidence in that kind of audit.
Why It Matters for Brazilian Agencies and SMBs
For agencies processing customer conversations, lead data or service information through the OpenAI API, the practical step is to assess today whether any production workflow handles personal data sensitive enough to justify requesting ZDR through OpenAI's sales team, especially on enterprise accounts already handling CRM data, service history, or end customer health and financial information. It is also worth documenting, for LGPD compliance purposes, which account endpoints operate under standard retention and which already sit under ZDR, since the ANPD has started demanding exactly this kind of documentary evidence, not just a generic privacy policy published on a website.