Major corporations are restricting the use of Anthropic's Fable model due to concerns over data retention policies. In June, Anthropic announced it would store usage logs for 30 days to mitigate security threats, prompting pushback from enterprise clients who fear for their intellectual property.
Nvidia, an investor in Anthropic, has limited Fable to non-sensitive tasks and relies on its own Nemotron models for internal operations like supply chain monitoring. Booz Allen Hamilton has banned employees from using Fable for cybersecurity work, citing risks of the model learning from proprietary code.
Palantir is blocking Fable deployment for its customers until Anthropic provides irrevocable zero-data-retention guarantees. Following customer pressure and similar moves by OpenAI, Anthropic plans to roll out a zero-data-retention program to select customers this fall.
Concerns persist regarding metadata collection and training practices. Even with zero-retention policies, labs collect technical usage data. Experts note that de-identification is often insufficient to prevent intellectual property leakage, as statistical patterns can be extracted from user interactions.
The issue of trust was highlighted by a case involving mathematician Tristan Buckmaster, who alleged that OpenAI used his research drafts uploaded to Codex to achieve a breakthrough. While OpenAI concluded after an investigation that the data did not influence its models, the incident underscored industry-wide skepticism.
