Add as a preferred source on Google

Ollama Cofounder Urges Companies to Set Clear AI Data Rules

Jeffrey Morgan said businesses should check where AI data goes, who can access it and how long providers retain it. He also described cost savings at two Ollama customers.

A hand disconnects a cable beside a compact computing workstation / TokenPost.ai
A hand disconnects a cable beside a compact computing workstation / TokenPost.ai

Ollama CEO and cofounder Jeffrey Morgan said companies using AI should examine how providers handle sensitive data as AI tools carry source code, internal documents and customer details into more business workflows.

In a Nov. 19 commentary, Morgan urged businesses to ask providers directly about data retention, get the answers in writing and make zero retention the default.

Provider terms vary, Morgan said. Some services may use customer data to train models, while others promise zero retention. Providers between those approaches may say they will not train on customer data but still keep it for a period of time.

Morgan also urged companies to understand where models run and which parties can access data. Open models can give businesses more control because they can run them in their own environments or through a trusted provider in a region approved by compliance teams. He said security teams should have another option if access to frontier tools is gated.

About 10 million developers worldwide use Ollama each month to run open models, Morgan said. He also said 85% of Fortune 500 companies have Ollama installed on desktops for local model use.

Morgan described cost figures from Ollama customers. One fintech company told Ollama it had spent 60% of its annual budget three months into a frontier-model contract, with 85% of the spending coming from engineers defaulting to the most expensive models. After two weeks of routing most engineering work to open models while keeping frontier models for complex problems, the company saw an immediate 30% savings on token costs, he said.

A large consumer AI company spending more than $10 million on frontier models found in its own human evaluation that Gemma 4 matched the model it had been using at roughly one-tenth the cost, Morgan said of an Ollama customer.

Morgan said some companies that had been wary of open models now use them, including models developed outside the United States, when they run on U.S. or European infrastructure under strict data terms.

Loading…