A new White House AI memo is expected to set clearer limits on how intelligence and national security agencies use artificial intelligence. According to The New York Times, President Joe Biden on Thursday is expected to sign a memorandum that addresses both the use of AI inside government and the protection of AI work from foreign adversaries.
The memo focuses on a central tension in national security AI: agencies may want powerful tools, but some decisions still require human judgment. The order points to areas where AI should not be allowed to act without human review, especially when the outcome could affect rights, surveillance, or security classification.
What the memo tells agencies to do
The memorandum is aimed at intelligence and national security agencies, including the Pentagon. Its core message is that AI can be used, but it must be surrounded by guardrails.
One of the clearest requirements is that humans stay in the loop for AI tools that may be used as targeting weapons. That matters because targeting decisions sit at the highest-risk end of AI use. The memo does not present AI as a replacement for human oversight in that context.
The order also bars AI from making several sensitive determinations without human review. Based on the source article, those include decisions about granting asylum, tracking someone based on their ethnicity or religion, and classifying a person as a known terrorist.
These examples show the kind of boundary the White House is trying to draw. The memo does not simply discuss AI in broad terms. It identifies specific uses where a machine-generated decision cannot be the final step on its own.
Why human review is the main safeguard
The memo’s emphasis on human review reflects a practical concern: AI systems can support decisions, but the most consequential uses still need accountable people involved. In the areas named by the order, a decision can carry major consequences for an individual or for national security operations.
Keeping humans in the loop does not mean agencies cannot use AI tools. It means the tools must operate within a process where people review and control important outcomes. That is especially important when an AI system might be used in targeting, screening, tracking, or classification.
The source article highlights three categories where the memo blocks AI from acting alone:
- Granting asylum decisions without human review.
- Tracking someone based on ethnicity or religion.
- Classifying a person as a known terrorist without human review.
Those limits are narrow but significant. They show that the White House is not only thinking about model performance or agency efficiency. It is also trying to define where automated systems should not make final decisions.
Protecting AI from foreign adversaries
The memorandum goes beyond how agencies use AI. It also calls for intelligence agencies to begin protecting work on AI and AI chips from spying or theft by foreign adversaries.
That part of the order frames AI development as a national security asset. If AI tools and AI chips are important to government capability, then the work behind them becomes something adversaries may try to obtain. The memo therefore treats protection of AI research and hardware-related work as part of the broader security challenge.
The source article does not list specific foreign adversaries or describe individual incidents. It does, however, make clear that the White House wants intelligence agencies to guard AI and AI chip work against espionage or theft.
This adds another layer to the memo. It is not only about preventing dangerous or unfair AI use inside government. It is also about preventing hostile nations or other adversaries from gaining access to sensitive AI capabilities before or after they are released.
The AI Safety Institute gets a role
The memorandum also empowers the recently established AI Safety Institute to help inspect AI tools before they are released. The purpose, according to the source article, is to ensure that those tools cannot aid terrorist groups or hostile nations.
That inspection role connects safety testing with national security concerns. Rather than focusing only on whether an AI tool works as intended, the memo points to the possibility that a tool could be misused by dangerous actors.
The article does not describe the inspection process in detail. Still, the role assigned to the AI Safety Institute suggests that the White House wants some form of review before certain AI tools reach the public or broader use.
In plain terms, the memo puts responsibility on agencies and safety reviewers before release, not only after problems appear. That is consistent with the broader theme of the order: AI should be managed before it creates serious risk.
How much impact the memo may have
The source article also notes an important uncertainty. As The New York Times points out, it is unclear how impactful the order will ultimately be because most of the deadlines it sets will lapse after Biden leaves office.
That timing issue matters. A memorandum can set expectations and assign work, but its practical effect depends on follow-through. If key deadlines fall after a change in administration, the long-term outcome may depend on whether the next government continues, changes, or slows the effort.
For now, the memo signals how the White House wants national security agencies to approach AI: use it with guardrails, keep people involved in sensitive decisions, protect AI work from foreign adversaries, and inspect tools before they can help terrorist groups or hostile nations.
The result is a policy step that treats AI as both a powerful government tool and a target that needs protection. Its ultimate force, however, may depend less on the text of the memo than on what happens after its deadlines arrive.