The intelligence was alarming. Ukraine had made a secret deal with Germany and France to deploy 5,000 to 10,000 soldiers, without insignia, to Greenland to thwart an American takeover. Denmark was re-deploying F-16 fighter jets to Greenland previously allocated to Ukraine. And in the United States, 400 Ukrainian refugees were arrested for violent protests over Greenland. Compounding concerns, this intelligence was leaked to major news outlets.
None of this actually happened. But what’s now called the Fake War For Greenland did. Legitimate-looking videos created with AI, complete with what appeared to be trustworthy media branding, flooded social media. SecAlliance, a cyber threat intelligence service, estimates 3.03 million viewed the disinformation campaign attributed to Russia during the height of the Greenland crisis with Denmark.
No wonder then that the new 2026 Annual Threat Assessment of the U.S. Intelligence Community says “AI is the defining technology for the 21st century.” The report acknowledges AI’s rising influence in the intelligence domain. “AI allows analysts to make sense of immense datasets and generate novel ideas and insights on complex national security issues.”
But there is an AI caveat. “These applications, however, also carry risks that require careful human engineering to appropriately mitigate risk of AI autonomy before they are broadly deployed.” Furthermore, AI is at the center of an emerging threat as adversaries weaponize AI to boost military power, cyber capabilities and global influence.
AI-generated realistic videos have the power to spread false narratives and create confusion. And that’s a big worry for the intelligence community. Questions about the reliability of AI data may, in fact, spur a comeback for old spy school techniques. Human-to-human intelligence gathering employing dead drops, brush passes, and in-person meetings may become more important as it becomes harder to distinguish between authentic and synthetic data.
A recent article in Studies in Intelligence, a CIA-backed academic journal, argues that AI’s ability to generate convincing deepfakes and fabricate messages introduces a new level of “noise” into digital communications. That means the value of communication methods that are not mediated through electronic methods is more valuable. This may be especially critical when the line between what is real and what is fake is blurred during a crisis when a quick decision is required.
“If my friend tells me face-to-face that he is in trouble, I can be confident that that’s true,” writes Thomas Mulligan, a RAND researcher and former CIA officer. “But when the same message is delivered through an electronic medium, it becomes more likely a scam than a bona fide plea for help.”
Intelligence agencies nonetheless are betting heavily on AI as the technology is increasingly being integrated into intelligence operations like that of the Central Intelligence Agency (CIA). The future is one in which teams of CIA officers manage teams of AI agents, said Israel Soong, deputy director of the CIA’s Office of AI, during a recent webinar with the Federal News Network.
“Imagine this in the future,” said Soong. “AI agents have already triaged the overnight intelligence for you. They’ve highlighted the most relevant operational development and they’re also combining and collating the intelligence reporting for you to review as you drink your morning coffee.”
Soong paints a rosy picture but there is growing concern that the rapid adoption of the Large Language Models (LLMs) that underpin AI may weaken critical thinking. A study conducted by Trends in Cognitive Sciences indicates the LLMs may standardize how users think and reason, thereby reducing cognitive diversity. Studies from Princeton, Wharton, and the University of Southern California indicate that AI may promote homogenized thinking and discourage intuitive insight and non-linear methods of critical analysis. The studies also revealed a tendency to rely on AI’s judgement even when users know it is wrong, a phenomenon called “cognitive surrender.”
“The more you use AI, the more you will use your brain in a different way,” said French admiral Pierre Vandier, NATO Supreme Commander for Transformation in an interview with Defense One. “And so we need to be able to have some oversight, to be able to critique what we see from AI and to be sure you’re not fooled by a sort of false presentation of things. It’s something we need to take care of.”
Perhaps most alarming from an intelligence standpoint is the finding from the Princeton study regarding “sycophantic AI” that instills a false sense of confidence and reinforces preconceived bias. “Our results show that default interactions of a popular chatbot resemble the effects of providing people with confirmatory evidence, increasing confidence but bringing them no closer to the truth.”
It’s enough to make spies work on their old school dead drop skills.
