NSA wants access to 'all the models,' deputy director says
The deputy director confirmed active talks with frontier model companies but offered no evidence the voluntary pre-release arrangement is operational.
TL;DR
NSA Deputy Director Tim Kosiba said Thursday the agency wants access to "all the models" and is holding extensive conversations with leading AI developers. The remarks are one of the clearest public indications of NSA's central role under a June executive order that encourages developers to voluntarily give the government 30 days of pre-release access for classified security testing. Kosiba didn't name the companies involved or confirm any developer has handed over an unreleased model.
The National Security Agency is in active talks with leading AI developers as it helps implement the testing framework called for by a June White House executive order, the agency's deputy director said Thursday. But his remarks didn't establish that any company has actually handed over an unreleased model.
"We want access to all the models, and we're going to take advantage of that," Tim Kosiba said at an Intelligence and National Security event in Bethesda, Maryland. "There are a lot of conversations happening amongst industry, especially the frontier model companies." He didn't name the companies involved, say which models NSA currently uses, or confirm any developer has provided pre-release access.
The remarks are one of the clearest public indications yet of how NSA is approaching its central role under the June order, which encourages developers to voluntarily give the government 30 days of pre-public access to advanced models for classified security testing. The order explicitly prohibits a mandatory licensing or approval process. NSA's director is responsible for deciding which systems qualify as a "covered frontier model," in consultation with ONCD and CISA.
A related national security presidential memorandum, also signed in June, directed intelligence and defense agencies to build "deep, proactive" industry partnerships and warned against becoming dependent on a single supplier.
Kosiba's comments come as government access to advanced AI continues to expand. OpenAI's GPT-5.6 models became available to federal customers through its FedRAMP-authorized enterprise service this month. The White House recently told major developers that open-weight models, whose underlying settings can be downloaded and modified, wouldn't be included in the testing arrangement. Separately, the Pentagon remains in a legal fight with Anthropic after designating the company a supply-chain risk over restrictions on military use of its models; a federal judge temporarily blocked parts of the government's action.
Published ·Deep Fathom