Nvidia and Booz Allen Hamilton restrict their use of Fable due to a lack of ZDR assurances; source: Palantir hasn't made Fable available via its own software

If you use a proprietary LLM for sensitive work, its data retention policies now present a concrete barrier to adoption.

Nvidia and Booz Allen Hamilton have reportedly restricted their use of Fable, a large language model (LLM) platform, due to concerns about Zero Data Retention (ZDR) assurances. The companies are seeking guarantees that their proprietary data will not be used for training or retained by the platform. This development surfaces amid broader industry anxiety about the potential for LLMs to inadvertently expose sensitive customer information or intellectual property. A source indicated that Palantir, the developer of Fable, has not made the platform available through its own enterprise software, potentially exacerbating these data privacy concerns for large clients.

The hesitancy of major technology and consulting firms like Nvidia and Booz Allen Hamilton to fully adopt Fable highlights a critical bottleneck in enterprise AI adoption: trust in data privacy. The demand for robust Zero Data Retention (ZDR) policies signifies that clients are moving beyond the capabilities of LLMs to focus on the security and confidentiality of their most sensitive information. This reluctance suggests that even powerful AI tools may remain underutilized if they cannot meet stringent data governance requirements, pushing vendors to prioritize security features over raw performance.

This situation signals a growing market demand for AI solutions that offer verifiable data isolation and privacy controls. Companies are likely to favor platforms that can demonstrate a clear separation of customer data from training sets and provide auditable proof of deletion or non-retention. As more firms express similar concerns, the competitive landscape for enterprise LLMs may shift towards those that can offer the highest level of data security and compliance, potentially creating a disadvantage for platforms unable to offer such assurances or those perceived as less transparent in their data handling practices.