US State Department Outlines AI Expansion Strategy in New Enterprise Playbook

The US Department of State has published a new Generative AI Playbook, publicly setting out how it built and scaled its internal AI chatbot, StateChat, across nearly its entire diplomatic workforce.

The document, dated July 2026, is pitched as a blueprint other federal agencies can follow as Washington pushes to embed AI more deeply into government operations.

According to the playbook, StateChat now has more than 62,000 users spanning 98 per cent of US diplomatic posts worldwide. The document traces the multi-year journey behind that expansion, from early staff experimentation with public chatbots in 2022 through to plans for autonomous AI agents that could soon draft cables without prompting.

How 150 Testers Became 62,000 Users

The department's Chief Information Officer, Dr Kelly Fletcher, and Acting Chief Data and Artificial Intelligence Officer Amy Ritualo wrote in the playbook that StateChat became the department's 'first Sensitive But Unclassified (SBU)-enabled enterprise GenAI chatbot,' built with support from the General Services Administration's Technology Modernisation Fund.

The rollout followed a staged path: 150 testers in the early 'alpha' phase, 8,000 in beta, and more than 62,000 by June 2026, according to the document.

The One-Year Security Hurdle That Unlocked StateChat

Handling sensitive diplomatic material required extensive security clearance. The playbook states that securing a 'FISMA High' Authority to Operate, the government's highest security classification for such systems, took roughly a year of work with cybersecurity and governance teams.

That clearance ultimately allowed StateChat to draw on sensitive but unclassified sources including internal cables, SharePoint files and official reports, expanding what the tool could actually do for diplomats day to day.

AI Agents That Work Without Prompts

The playbook's final chapter signals where the department wants to go next. It describes plans to move 'from tools users must explicitly activate to agents that work together autonomously', citing an internal tool called FAMsearch, which already answers policy questions by pulling from the department's 30,000-page Foreign Affairs Manual without staff opening a separate application.

The document imagines a future where a diplomat drafting a cable about visa policy could have one agent automatically flag relevant guidance, another check related cables, and a third suggest formatting, all without being separately requested.

The Users Who Did Not Save Time and Why

The playbook is not entirely celebratory. A department-run impact survey found that 90 per cent of regular users reported saving time, with users saving an average of 1.6 hours per week per task.

However, the same survey found that among users who said StateChat did not save them time, the top complaints were 'insufficient quality of responses' and issues where staff spent more time verifying outputs than they would have spent doing the task manually.

The document also notes that some early enthusiastic adopters disengaged over time due to 'accumulated small barriers', including promised features that failed to materialise.

The State Department's playbook lands as generative AI use across the federal government accelerates sharply. A Government Accountability Office report found that federal agencies' AI use cases nearly doubled between 2023 and 2024, with generative AI use climbing ninefold in the same period, and separate GSA figures show that programme was in part fuelled by an $18.2 million Technology Modernisation Fund award specifically for the State Department's generative AI diplomacy work, according to Brookings Institution research published in April 2026.

That same research notes AI adoption across government remains uneven, hampered by a shortage of AI-specialised staff, a risk-averse culture and public scepticism about how the technology is used. The State Department's decision to publish its playbook publicly, rather than keep the lessons internal, suggests officials see wider federal AI adoption, and public confidence in it, as unfinished business that other agencies will need help navigating.