CIA to embed AI co‑workers
The CIA plans to embed AI “co‑workers” across its analytic platforms over the next few years, signalling a move from pilots to platform-wide integration in sensitive institutions. That normalisation suggests heavily supervised, instrumented AI inside existing systems is becoming institutionally acceptable — with significant implications for governance and procurement patterns. (politico.com)
The Central Intelligence Agency has already used artificial intelligence to produce its first autonomous intelligence report, and Deputy Director Michael Ellis said on April 9 that “AI co-workers” will be built into all of the agency’s analytic platforms within the next few years. Those systems are being aimed at the part of the agency that turns scraps from spies, sensors, and intercepted data into judgments for presidents and cabinet officials. (politico.com) These are not pitched as robot bosses. Ellis said the tools will help with basic analytic work like drafting key judgments, testing conclusions, and spotting trends, while “human beings” remain the ones making key decisions. (politico.com) That is a bigger shift than it sounds, because intelligence analysis is the Central Intelligence Agency’s oldest core job. Since 1947, the agency has been the analytic hub of the United States intelligence community, which means its reports are supposed to be the cleaned-up final product after raw clues come in from many places. (politico.com) The pressure pushing this is volume. The Office of the Director of National Intelligence and the Central Intelligence Agency said in their March 8, 2024 open-source intelligence strategy that publicly and commercially available information had become so expansive and fast-changing that the community needed to modernize how it collects, creates, and delivers analysis. (odni.gov) Inside the agency, that modernization job already has a home. The Directorate of Digital Innovation, created in October 2015, says one of its five pillars is to “Harness the Power of Data and Artificial Intelligence,” alongside cyber, partnerships, open source, and innovation. (cia.gov) The Central Intelligence Agency was not starting from zero this week. Nextgov reported that the agency managed a few hundred artificial intelligence projects last year, which means the new announcement is less a lab experiment and more a decision to wire those tools into everyday analytic software. (nextgov.com) Agency officials have been describing the same model for months: speed from machines, judgment from humans. Chief Artificial Intelligence Officer Lakshmi Raman said in June 2025 that artificial intelligence can accelerate processing and automate tasks, but risk decisions, intent judgments, and final calls still have to stay with trained people. (govciomedia.com) The procurement side is moving in parallel. On February 9, 2026, the Central Intelligence Agency announced a new acquisition framework with centralized vendor vetting and a faster information-technology authorization process so it can bring in “breakthrough technology prototypes” and modernize core systems more quickly. (nextgov.com) Ellis also signaled that the agency does not want private model makers setting the outer limits of how intelligence officers use these systems. Politico reported that he said the Central Intelligence Agency would “not let private companies dictate how and when the CIA will make lawful use of their technologies,” a line that landed amid a Pentagon fight with Anthropic over contractual guardrails. (politico.com) So the news is not just that one spy agency likes artificial intelligence. It is that one of the most secretive and procedure-heavy parts of the United States government is moving from pilots to platform-wide deployment, with humans supervising the output and procurement systems being rebuilt to buy more of it faster. (politico.com) (nextgov.com)
