US National Security Agency Wants Access to All Commercial AI Models
The NSA is expanding cooperation with AI developers to enhance cyber defense capabilities and test national security risks.
US National Security Agency (NSA) Deputy Director Tim Kosiba announced that they are conducting talks with leading developers to implement new White House directives and want to gain access to all commercial AI models on the market.
New Directives and the Quest for Access
NSA Deputy Director Tim Kosiba, speaking at an event in Maryland, stated that the agency has been using AI for a long time, but they are adopting a more aggressive approach due to the rapidly evolving capabilities of new systems. Kosiba expressed that they want to access all models in the commercial market and leverage these technologies.
This move comes after a White House executive order issued in June, which tasks national security agencies with testing the advanced hacking capabilities of AI systems. The executive order includes a voluntary mechanism requiring developers to submit their models for government testing for up to 30 days before releasing them to the market.
Relations with Companies and Limitations
The NSA's access request can cover deploying models in a government environment or subjecting them to security testing. While OpenAI's GPT-5.6 models were made available to federal customers this month, open-source models are excluded from this testing program, unlike the closed systems of companies such as Google, Anthropic, and OpenAI.
On the other hand, a legal dispute is ongoing between the US Department of Defense and the firm Anthropic due to military use restrictions. Despite the Pentagon defining the firm as a supply chain risk, the NSA has special agreements with Anthropic, and the agency is experimentally using the company's advanced Mythos model to test the defense of military networks.
Legal Compliance and Human Control
As the cyber defense and attack detection capabilities of AI models develop, the risks of misuse of these technologies are also being discussed. Kosiba emphasized that in the process of accelerating AI use, the compliance of the obtained results with surveillance laws and internal rules will always be overseen by humans.