Goldman Sachs Chief Information Officer Marco Argenti (마르코 아르젠티) said companies have no reason to exclude open-weight models if they put certain safeguards in place, Axios reported on Sept. 8 (local time).

Argenti said, "If you install safety devices, companies including banks can use any model from any company with confidence." He said, "I do not think access itself should be blocked at the outset for any model. We should ask what conditions allow any model to run, including Chinese models."

Most or all of the open models Goldman Sachs currently uses are made in the United States. Goldman Sachs has tested several U.S.-made open models, and some are used in actual work.

The Trump administration has not yet disclosed details of a voluntary framework on AI self-regulation, but open models will be excluded, Axios reported, citing a source.

U.S. Treasury Secretary Scott Bessent (스콧 베센트) and Michael Kratsios (마이클 크라시오스) at the White House Office of Science and Technology Policy recently posted a piece supporting U.S.-made open-source AI.

Argenti applied two concepts used in cybersecurity directly to AI models. One is "zero trust," which assumes the model may already have been breached. The other is "defense in depth," which places multiple independent defensive barriers between the model and the company's systems.

This means the AI system cannot directly access source data and instead must go through a bank data platform with access controls applied.

Argenti said, "I want to have options without increasing risk." He said, "If you have sufficient defensive barriers, a company can run virtually any model without greatly increasing risk."