CIA rolls out AI assistants

The Central Intelligence Agency has begun deploying generative‑AI assistants inside its analytic platforms while insisting humans will retain final decision authority. (politico.com) Deputy Director Michael Ellis announced the rollout on April 9, 2026 at a Special Competitive Studies Project event and said “within the next couple of years” AI “co‑workers” will be built into analytic tools. (politico.com) Ellis told reporters the agency ran roughly 300 artificial‑intelligence projects last year and that it recently used AI to produce its first fully autonomous intelligence report, without naming the model or its subject. (nextgov.com) (nextgov.com ) Agency officials framed the push as a way to speed analysis and keep pace with rivals, echoing broader CIA priorities to maintain a “decisive technological advantage” over competitors such as China. (politico.com) Ellis said the assistants will draft judgments, test analytic conclusions, flag trends and triage large data sets while human analysts remain in the loop for final judgments. (nextgov.com) (nextgov.com ) Ellis also warned the CIA will “not let private companies dictate how and when the CIA will make lawful use of their technologies,” an apparent swipe at recent Pentagon disputes with Anthropic over model guardrails and a March‑April 2026 supply‑chain risk designation. (yahoo.com) Outside experts and civil‑liberties groups have urged limits, saying AI‑assisted intelligence risks amplifying bias, producing “hallucinations,” and expanding surveillance without safeguards; organizations including the Brennan Center and the American Civil Liberties Union have called for guardrails. (brennancenter.org) Ellis characterized the work as an initial phase and said the agency plans to scale classified generative‑AI tools across analytic platforms in the coming years while keeping human oversight for final decisions. (nextgov.com) (nextgov.com )