US AI NSM Primer, Oct 2024
As a companion to muckrAIkers sixth episode, US National Security Memorandum on AI, Oct 2024, we wanted to release a short blogpost summarizing key takeaways from the lengthy document. Perhaps it will be moot in a couple months, but we can still use it to gain insights on how the US government is addressing “AI”.
A Behemoth Shouldn’t Twitch
When large institutions make fundamental changes, the repercussions have potential to be far reaching and unpredictable. The United States government is such an organization, with the military industrial complex budget set at approximately 850 Billion USD for the fiscal year of 2024 1. While the slow speed of the government is a source of great frustration to many, including myself, it in general increases institutional stability, and decreases risk [2,3].
The US National Security Memorandum on AI published on October 24, 2024, along with a framework on using AI for national security exemplifies this well; it is a continuation of last year’s Executive Order on AI, with a specific eye towards national security. Similarly to the first installment, the Executive Branch sets many explicit deadlines for the teams and initiatives that were kicked off last year, and also puts forth a number of directives.
To us muckrAIkers, this memorandum proposes an opportunity to understand how the Biden administration is thinking about AI.
Themes
Safe, secure, and trustworthy AI. This phrase is repeated frequently throughout the memorandum and framework. We see this as indication that the Biden administration understands these three characteristics are not necessarily the default when it comes to AI.
Promotion of democratic values. While it may seem obvious that a document from the President of the United States explicitly calls out the importance of democratic values, but maybe not…
Utilization of AI. Government entities are tasked with the incorporation of AI systems into their national security relevant procedures.
Prohibited AI Uses
- No targeting, profiling, or tracking people solely for exercising constitutional rights like free speech
- No interfering with free speech, or access to legal representation
- No unlawful discrimination
- No unlawful sentiment analysis
- No use of only biometric data for profiling
- No use for military estimates (noncombatant identification) unless there is sufficient testing + assurances, and trained human oversight
- No final determination of immigration classification (asylum/US entry)
- No production of reports based solely on AI outputs, unless there is a disclaimer
- No removal of human oversight from presidential decisions to use nuclear weapons
Deliverable Timeline
| Date | Deliverable |
|---|---|
| Nov 23, 2024 (30) | Establish a formal working group to guide how artificial intelligence is purchased and used for defense and security purposes, with special attention to protecting national security systems. |
| Dec 8, 2024 (45) | Establish an AI National Security Coordination Group, which will then create and maintain guidelines on how AI systems are developed, bought, and used for national security. |
| Jan 22, 2025 (90) | Make visas for people with sensitive technological backgrounds easier to get. |
| Review identification and evaluation of foreign threats to US AI dominance (and chips). | |
| Coordination Group makes Talent Committee (by deadline), which will then create government standards for finding, hiring, and keeping AI professionals | |
| Feb 21, 2025 (120) | Many departments shall each create new training programs and educational opportunities to help their employees gain AI knowledge and skills. Also includes new hiring highly skilled individuals. |
| Rapid systemic classified testing of AI models capabilities on (a) cyber threats; (b) nuclear safety risks, both public and classified – be able to move models to classified facilities if necessary | |
| Strategy to work with other countries to create shared rules/standards for safe AI development (which the US likes) – AI governance norm co-development w/ allies. | |
| Mar 23, 2025 (150) | Evaluation of feasibility of promoting co-development of AI with allies (other countries). |
| Issue cybersecurity guidance and/or direction for all AI used for national security | |
| Apr 22, 2025 (180) | Analysis of AI talent market – US + worldwide |
| Competitive advantage analysis of US private sector and how to maintain it Includes chips design + manufacturing; capital; specialists; compute+energy | |
| Launch project to assess feasibility of the federal government making a frontier model | |
| Threat analysis of the AI supply chain | |
| Guidance for AI developers on how to test and manage risks relating to safety, security, and trustworthiness | |
| Development of recommended benchmarks or other assessments of AI system capabilities and limitations | |
| Be able to rapidly test nuclear threat level of a model (within 30 days), and actually do it | |
| Start gathering the individuals to plan voluntary best practices with regards to biochemical technologies. | |
| All agencies must update their policies and procedures to explicitly include AI, these will apply to all contractors/sub-agencies. | |
| Subject to private sector cooperation—Voluntary preliminary testing of at least two frontier AI models prior to public deployment on harmful capabilities | |
| May 22, 2025 (210) | Roadmap for future classified evaluations of biochemical threats exacerbated by AI |
| Written recommendations on changing existing regulations/guidance to promote creation of AI for US national security purposes | |
| Jun 21, 2025 (240) | Use AI to enhance biosafety and biosecurity |
| Jul 21, 2025 (270) | AISI report to president including summary of AI safety findings, summary of necessary risk mitigation, adequacy statement about the tools/methods used to reach those conclusions |
| DOE report to president on nuclear threat, recommendation of corrective action, adequacy statement about the tools/methods used to reach those conclusions | |
| Pilot project to conduct classified tests on biochem capabilities | |
| Report on activities relating to the memorandum | |
| Oct 24, 2025 (365) | Joint report on consolidation and interoperability of AI efforts and systems pertaining to national security |
| Apr 17, 2026 (540) | Guidance on promoting benefits and mitigating risks of in silico biochem research |
Additional Items of Note
Chips are not neglected. The importance of advanced computer chips and the chip supply chain has not gone unnoticed.
Talent is emphasized. To maintain it’s edge, the US will need to remain the go-to destination for technical expertise relevant to AI.
Presidential reports. Many groups will be tasked with the preparation of annual reports, to be given directly to the president.
Blocking foreign acquisition. One item explicitly states that the United States government may block the sale of an AI company to foreign parties to prevent leakage of items instrumentally useful to the creation and effective use of powerful AI systems.
