OpenAI is gaining enterprise work while Anthropic customers wait for zero-retention access

Anthropic’s data-retention dispute is producing a measurable commercial consequence. Major customers are still waiting for zero-data-retention access to the newest Fable models, and some are expanding their use of OpenAI in the meantime. Palantir and Booz Allen are using more of OpenAI’s Astra model, The Information reports.

This is a narrower claim than saying Anthropic trains on customer secrets. The issue is whether prompts, outputs and usage records are stored at all, including for safety monitoring. A customer handling classified, proprietary or regulated material may require irrevocable deletion even when the provider promises never to use the material for training.

The episode shows how model competition changes in sensitive enterprises. Benchmark quality can be outweighed by a deployment property that ordinary users rarely see: which model versions support which retention contract, through which product tier and technical path. Anthropic may close the gap by extending zero-retention support, but during the delay OpenAI gains workload, integration experience and internal advocates that can be difficult to dislodge later.