Nvidia Corp. (NASDAQ:NVDA), Palantir Technologies Inc. (NASDAQ:PLTR) and Booz Allen Hamilton Holding Corp. (NYSE:BAH) are reportedly tightening their use of advanced AI models over concerns about exposing sensitive corporate information.
The companies are demanding stronger guarantees or dropping some advanced models from Anthropic and OpenAI, The Information reported Monday.
The concern is that model providers could retain or learn from customer data and proprietary knowledge.
Nvidia shares fell about 3% Monday amid a broader selloff in AI-linked stocks.
Friedberg Says Ohalo Is Moving Off Claude
All-In co-host David Friedberg, CEO of agricultural biotechnology company Ohalo, described a similar concern on Friday’s podcast.
Friedberg said Ohalo is "moving off Claude for a lot of our sensitive data" and shifting work toward locally run models.
He also described several anecdotal cases where novel scientific ideas discussed with one model appeared to resurface when he later queried another version from a different account.
Friedberg said his concern is not just leaking confidential documents, but potentially giving AI providers valuable ideas and know-how developed inside a company.
Nvidia and Palantir Build an Alternative
Nvidia is already taking a similar approach with one of its most sensitive operations: its supply chain.
Last week, Nvidia and Palantir unveiled a system that uses Palantir software and Nvidia’s open Nemotron models to analyze Nvidia’s own supply-chain data. Nvidia is the first customer for the technology.
Nvidia says companies can customize the models using proprietary operational data while keeping control of their data, models and where the system runs.
For Nvidia, that means using AI to capture how its planners make decisions without handing that supply-chain knowledge to an outside model provider.
Anthropic Responds to Customer Pressure
Anthropic introduced 30-day data retention with Fable 5 in June to detect misuse across multiple sessions and accounts. The company says that information is not used to train its models.
Anthropic later acknowledged that many enterprises, particularly in regulated industries, found the retention requirement difficult to accept.
Its proposed solution, Enterprise Frontier Safeguards, will keep activity data used for misuse monitoring in customer-controlled cloud infrastructure. It begins rolling out in phases later this fall.
The data concerns have not dented expectations for Anthropic’s IPO. Polymarket still puts the odds of a listing this year at 86%.
Nvidia is also reportedly considering a roughly $10 billion anchor investment in an offering that could value Anthropic near $2 trillion.
The debate echoes Palantir CEO Alex Karp‘s recent warning that companies risk handing AI providers their proprietary "alpha" along with their data.
Image: Shutterstock
Login to comment