The operational shift here is concrete: before June 2026, Nvidia and Palantir routed sensitive workloads through Anthropic’s flagship Fable model like any other enterprise customer. After Anthropic introduced a rolling 30-day retention window for Fable that stops short of an irrevocable zero-data-retention (ZDR) commitment, Nvidia moved its confidential work — including supply-chain analysis — onto its own Nemotron models running inside a Palantir-built system, reserving Fable for lower-stakes tasks like open-source development, according to The Information as reported by Reuters and detailed further by Tom’s Hardware and BiGGo. The cost was not hypothetical: a large U.S. utility company scrapped plans to test Fable on core power infrastructure after Anthropic wouldn’t budge on a nonrevocable ZDR policy.
What makes this worth tracking is the overlap between the companies raising the alarm and the companies positioned to profit from the alternative. Nvidia and Palantir aren’t neutral bystanders assessing Anthropic’s data practices — they’re also the vendors of Nemotron and the “sovereign AI stack” that customers get steered toward instead, an arrangement Yellow.com describes as already courting the roughly 2-million-employee U.S. civilian government workforce. Palantir CEO Alex Karp has framed the stakes in proprietary terms, warning publicly that companies risk handing over their competitive “alpha” along with their data.
Nvidia and Palantir’s own commercial stake in Nemotron and the sovereign-AI stack means their expressed data caution is also a sales pitch for the product they’d rather sell instead.
The pattern extends well past two companies. Booz Allen Hamilton has pulled back from certain frontier models on classified work, per BiGGo; Northrop Grumman has gone further, running open-source models on fully air-gapped servers; Novo Nordisk keeps using Claude but bans proprietary data inputs outright; and telecom C Spire says its no-training contracts with OpenAI and Anthropic still leave ambiguity around what usage metadata gets collected. Microsoft is capitalizing directly, pitching isolated cloud deployments that keep data off external servers entirely — at a price point BiGGo notes some customers are still weighing.
Anthropic’s answer, per BiGGo, is a proposed “Enterprise Frontier Safeguards” framework that would keep misuse-monitoring data inside customer-controlled infrastructure, with a phased rollout starting later this fall. Whether that closes the trust gap is the thing to watch: if it doesn’t include an irrevocable ZDR guarantee — the specific term Nvidia and the canceled utility customer both cited — expect more enterprise workloads to migrate toward open-weight, self-hosted setups, which is precisely the outcome Nvidia and Palantir have also been lobbying federal procurement rules to protect, per FedScoop’s coverage of the GSA’s draft AI acquisition clause.
Nvidia has also decided to use Fable for tasks that don't require it to gain access to sensitive data. The company points to the same lack of ZDR guarentees as the reason. Instead, Nvidia uses its own in-house AI solution for tasks that it deems too sensitive for Anthropic's model.
— Reuters
