Anthropic's Fable model has raised data trust concerns among major clients, prompting companies like Nvidia and Booz Allen Hamilton to restrict its use for sensitive work. The issue emerged after Anthropic announced it would retain usage logs for 30 days to defend against complex attacks.
Nvidia, an investor in Anthropic, has limited Fable's use to less sensitive tasks like open-source projects and now runs its own Nemotron models for internal work. "We believe ZDR should be on by default," said Justin Boitano, Nvidia's VP of Enterprise AI, according to The Information.
Booz Allen Hamilton, an early user of Anthropic's Mythos model, has banned Fable for work on proprietary cybersecurity software. CTO Bill Vass expressed concerns that Fable might learn from their code. Palantir is also blocking Fable deployment until Anthropic provides zero-data-retention guarantees, per The Information.
"Companies are tired of being exploited by AI labs," said Alex Karp, Palantir's CEO, at a customer event. Palantir's stance is partly self-serving, as the company wants customers to use its secure platform instead of direct AI providers.
In response to customer pushback and OpenAI's August move to let GPT-5.6 Cyber customers store logs on their own servers, Anthropic is rolling out a similar program to select customers this fall. However, zero data retention still leaves gaps, as both OpenAI and Anthropic collect metadata and technical usage data from enterprise customers.
"De-identification is weak," said John Schulman, OpenAI co-founder. "Users can be traced back with just a small number of bits." AI researcher Sarah Hooker warned companies that even synthetic data techniques can preserve privacy while extracting statistical patterns from user data.
Source: thedecoder
