OpenAI Log Preservation Order Shows Agent Builders Don't Control Data Retention

srivas_ashi · reddit · 2026-10-29

The author uses the 2025 NYT copyright case, where a federal magistrate ordered OpenAI to preserve output logs that would otherwise be deleted (including deleted chats, Temporary Chats and API content), to argue that agent builders calling public LLMs may not control how long their data is retained.

The blanket order ended Sept 26, 2025, but already-preserved logs stayed preserved (except EEA, Switzerland and UK users); in January 2026 a judge affirmed an order requiring OpenAI to produce a sample of 20 million de-identified logs.
OpenAI's stated policy is to remove deleted data within 30 days unless legally required to retain it.
Impact on agent builders: agents touch CRM records, documents, tickets and customer messages, sending them to a public LLM on every call; if the provider is under legal hold, retention is out of your hands.
Some teams therefore run agents on private models in a VPC or on-premise to control retention, access and audit, though they still must manage logs, memory stores and backups; public LLMs remain faster for low-stakes workflows.

The author asks three questions: whether provider retention terms are a design constraint, whether full prompts and tool outputs are logged with a purge policy, and whether sensitive-data clients use zero-retention public APIs or private models. The author discloses working in enterprise AI agents and private deployment.

Original post →

OpenAI Log Preservation Order Shows Agent Builders Don't Control Data Retention

More from coding & agent