Anthropic data fears prompt Nvidia (NVDA), Palantir (PLTR) and Booz Allen (BAH) to restrict model use, reports The Information

Restrictions of this kind, where large customers limit internal use of a model provider over data-handling concerns, sit in a now-familiar pattern around enterprise AI adoption: the binding constraint has repeatedly been governance and data leakage risk rather than model capability, and past episodes of this type have tended to resolve through negotiated deployment terms, on-premise or isolated instances, rather than outright abandonment. The actors matter here: defense-adjacent and government contractors operate under stricter data-security obligations than commercial enterprises, so their caution is partly structural and does not automatically read across to the broader customer base. The distinction worth drawing is between a pause pending security review and a procurement-level exclusion, since the former is common and reversible while the latter flows into revenue with a lag. Attribution to a single press report rather than company confirmation is itself a tell: episodes sourced this way have often been walked back or narrowed once the named firms comment. The follow-ons are on-the-record statements from the companies involved and any indication of whether the restrictions cover one provider or reflect a wider tightening of third-party model usage.