The CIA has produced its first intelligence report written entirely by an AI system, Deputy Director Michael Ellis confirmed on April 9 at a Special Competitive Studies Project event in Washington, D.C. The report emerged from one of more than 300 AI projects the agency ran last year.
Ellis didn't share what the report covered, which model generated it, or what safeguards were in place. Operational security, naturally. So we're left with a milestone that's simultaneously striking and impossible to evaluate.
What "AI coworkers" actually means
The near-term plan is less dramatic than the headline. Ellis described AI tools embedded in the CIA's analytic platforms to handle drafting, editing, and checking outputs against tradecraft standards. "It won't do the thinking for our analysts, but it will help draft key judgments, edit for clarity and compare drafts against tradecraft standards," he said, which sounds a lot like what every enterprise AI vendor promises, just classified.
Human analysts keep final sign-off. The interesting part is the ten-year vision: officers managing teams of AI agents as "autonomous mission partners." That's a significant leap from spell-checking intelligence briefs, and Ellis offered no detail on how the agency gets from here to there.
The Anthropic problem
There's a backdrop to all of this that Ellis danced around without naming directly. Anthropic declined to ease restrictions on its tools for domestic surveillance and autonomous weapons applications. Defense Secretary Pete Hegseth responded by designating the company's products a supply chain risk, and President Trump ordered federal agencies to phase out Anthropic's tools. The company is challenging the move in court.
Ellis's comment that the CIA "cannot allow the whims of a single company" to constrain its capabilities landed clearly enough. The agency is diversifying across vendors, which makes operational sense regardless of the political theater. Relying on a single AI provider for classified intelligence work was always a fragile arrangement.
Why China keeps coming up
Ellis framed urgency around Beijing. The CIA doubled its technology-focused foreign intelligence reporting, tracking how adversaries use AI in semiconductors, cloud computing, and R&D. "Five to ten years ago, China was nowhere near America, in terms of technological innovation," Ellis said. "That's just not true today."
The agency also elevated its Center for Cyber Intelligence to a full mission center, a structural change Ellis said is already producing results in deploying tools and gaining access to targets. Connecting that promotion directly to AI development and defense suggests the CIA sees cyber operations and AI capability as inseparable, which tracks with how the 2026 Worldwide Threat Assessment treated AI as a force multiplier across every threat category rather than a standalone concern.
So what's actually new here?
Strip away the event-speech framing, and the CIA is describing roughly the same AI integration path as every large organization: assistants first, agents later, humans in the loop until they're not. The difference is that 300 AI projects at Langley carry stakes that a Fortune 500 company's chatbot rollout doesn't.
The agency hasn't disclosed which models it uses, how it validates AI-generated intelligence, or what happens when an autonomous system flags a threat that turns out to be noise. Those are the questions that matter, and none of them got answered on April 9.
