The Biden administration is about to be out of office, and in its final months it has been furthering the work it started on cybersecurity and national security initiatives. The latest development, perhaps the last major one before the transition, is a national security memo that addresses the safe use of AI in military and intelligence applications.
Some of the memo is an exhortation to Congress to authorize funding for projects seen as vital to keeping up with China in these areas. But federal agencies can be expected to immediately pick up the pace in integrating AI and recruiting talent, and both protections and new risk guidelines will be coming soon for AI developers.
Federal agencies ordered to put new focus on AI
The memo stresses three AI-related points considered vital to national security: infrastructure, protection for domestic AI developers, and a strong recruiting pitch to the world’s AI talent. The administration can do little on that first point right now, in its waning days and with an election about to wrap up; the need there is mostly for ambitious new energy and data center projects, something Congress will have to authorize the funding for. But federal agencies have been directed to provide new security assistance to US AI developers, mostly in the interest of thwarting Chinese spies looking to steal technology, and to reform their hiring practices to emphasize AI talent acquisition.
In terms of governance, the agencies tasked with national security have also been ordered to designate a chief AI officer and have these new leaders form an AI National Security Coordination Group. The US AI Safety Institute is also taking on greater responsibility, serving as the point agency for interfacing with private companies on AI evaluation and testing. This agency has been given six months to produce new risk guidance for developers and will immediately begin preliminary testing of “frontier” AI models that have not yet been deployed to the public.
Thus far the public-private partnership seems to be focused on the developers most directly concerned with national security, those that are already likely looped in on cybersecurity issues. There is more expansion to be done to involve state and local stakeholders, as well as investors and those training tomorrow’s AI workforce. And safety and playing defense is far from the only concern, as the administration also emphasizes the need to be at the cutting edge of technological innovation.
National security memo tries to balance military innovation with safety
Even more of the 40-page national security memo is devoted to the need for government to adapt to AI’s anticipated prominence in warfare and espionage. China is the focus here, seen as the only US rival at present with the technological development to create equivalent parallel systems in this area.
The two countries have agreed to keep dialogue open about AI’s biggest safety risks, and had an initial round of talks this past May that seemed to go well. But beyond that it is open competition, and the US is leery of a “strategic surprise” integration of AI into the Chinese military.
The national security memo cautions that this does not mean the US can rush into military AI adoption, however. It specifically warns that AI should be kept away from automated military decision-making and lays down new rules for autonomous weapons to ensure safe and ethical use. The memo also addresses the common fear of AI gaining access to nuclear codes, warning that it must entirely be kept away from these systems.
