Reboot Weekly: What 250,000 Public Workers Taught Us About AI

Published on October 29, 2026

Summary

After training more than 250,000 government workers through InnovateUS, the first Public Sector AI Observatory report, as featured by Fast Company and the Boston Globe, finds that confidence drives adoption. Beth Simone Noveck looks at South Australia, where a teacher-built tool cut writing assessments from 30 minutes to 52 seconds and saved an estimated 22,544 hours statewide. Also this month: Neil Kleiman on why government is falling behind on AI adoption, 2,702 Californians on how AI should be governed, a seven-organization report on bossware in California workplaces, and Anthropic’s disclosure that Georgia voter data was accessed without authorization while staff were using Claude.

Upcoming InnovateUS Workshops

InnovateUS delivers no-cost, at-your-own-pace, and live learning on data, digital, innovation, and AI skills. Designed for civic and public sector, programs are free and open to all.

Upcoming InnovateUS series (free, live):

Worker-Centered AI Adoption in the Public Sector ›
For public-sector managers and leaders navigating AI adoption in their workplaces.

AI Insourcing and the Government Product Model › (new)
How agencies can move from buying technology to building digital products and services, including when to buy, when to build, and how to organize teams.

The Good, the Bad and the Ugly of Predictive AI ›
With Princeton’s Center for Information Technology Policy, this series asks when predictive systems improve public-sector decisions and when they introduce new risks.

[ See all workshops → ]

AI for Governance

What We Learned from 250,000 Public Workers About AI Adoption (Observatory of Public Sector AI: First Report)

Beth Simone Noveck, Gregory Porumbescu, and Lee Rainie on October 26, 2026 in Reboot Democracy (reprinted from Fast Company)

After training more than 250,000 government workers in all 50 states and DC through InnovateUS, the inaugural Public Sector AI Observatory report finds that confidence is what drives adoption: highly confident respondents planned to apply AI to an average of 3.8 tasks, versus 1.5 for low-confidence ones. The authors draw on exchanges with named practitioners — including New York City’s health commissioner on AI for benefits delivery and a Colorado state leader using generative AI to improve performance reviews, with reported gains in feedback quality, worker satisfaction, and reduced turnover — to show how peer learning turns individual experiments into shared practice.

Read article

AI for Governance

The AI Race Government Is Losing

Neil Kleiman on October 15, 2026 in Governing

Burnes Center senior fellow Neil Kleiman argues the US is winning on frontier models and labs but losing on the race that matters for citizens: public-sector adoption. Drawing on RethinkAI’s two years of research — mostly avoidance and uncoordinated experimentation — plus Code for America’s 2026 50-state assessment, he points out that state lawmakers have introduced 1,600+ AI bills since 2019, 77% focused on restricting or auditing use rather than enabling it. Meanwhile Beijing’s “AI Plus” initiative targets more than 70% adoption of AI devices and agents across government by next year.

Read article

AI and Education

Digital Education: What Schools Can Learn from Australia About Using AI

Beth Simone Noveck on October 27, 2026 in Reboot Democracy (adapted from Süddeutsche Zeitung, October 23, 2026)

Rebecca Weber, an English teacher at Adelaide Botanic High School, used to spend 30 minutes assessing each student’s writing against a 14-level English-proficiency scale; now she spends 52 seconds, thanks to a tool she built on EdChat, South Australia’s secure AI system. The state education department gave her time to develop the tool, tested it in 50 schools, trained teachers, and made it available statewide — reaching 58,000 writing samples and saving an estimated 22,544 hours by end of 2025. Dan Hart (former Head of AI, NSW Department of Education) describes building a custom orchestrator to retain data ownership and model flexibility. Noveck argues US schools can adopt the Australian model: give teachers time, test, scale what works.

Read article

AI and Public Engagement

Beth Simone Noveck Says AI Can Change Democracy for the Better — If We Choose to Use It That Way

Kevin Chang Barnum, Khalilah Brown-Dean, Erica McIntosh, Robyn Doyon-Aitken, and Megan Fitzgerald on October 16, 2026 in Connecticut Public (Disrupted)

Beth Simone Noveck, Director of the Burnes Center for Social Change at Northeastern University and former Deputy Chief Technology Officer at the White House, joins Disrupted to make the case from her new book Reboot: AI and the Race to Save Democracy: AI’s destructive potential is real, but so is its capacity to repair democratic institutions — if we choose to use it that way.

Read article

AI and Public Engagement

Humans in the Loop: Californians Say How AI Should Be Built, Used, and Governed

Engaged California staff on October 13, 2026 in Engaged California

Engaged California’s Humans in the Loop project asked Californians how AI is reshaping work and the economy — and what the state should do about it. 2,702 people took the survey; 103 joined facilitated Zoom discussions that refined gut reactions into 26 concrete policy ideas. The top three: (1) hold AI companies accountable for the harm they cause, (2) protect people and the environment from AI’s resource demands (data center noise, water, electricity), (3) set rules for where AI can be used safely, including requiring humans to make final judgments.

Read article

Governing AI

Opinion | What if A.I. Is Just a “Normal Technology”?

Ezra Klein and Rollin Hu on October 9, 2026 in The New York Times

In this Opinion podcast, Ezra Klein interviews computer scientist Arvind Narayanan, who argues that just because AI is intelligent doesn’t mean it’s powerful — pushing back on narratives of civilization-scale disruption and reframing AI as a “normal technology” that scales incrementally and governs itself through familiar institutional patterns.

Read article

Governing AI

The Promise and Peril of “AI for Democracy”

Dean Jackson on October 7, 2026 in Kettering Foundation

Dean Jackson tracks the shift from the deepfake-panic phase of the AI-and-democracy conversation to one centered on AI-facilitated deliberation. Drawing on a recent 23-scholar paper on AI’s impact on democracy, he discusses “bridging algorithms” and “collective dialogue systems” that help governments listen at scale, and cites tech-industry proposals (Sorota and Handler in MIT Technology Review) for AI mediators. The through-line: deployed thoughtfully, AI could lower barriers to participation; adopted carelessly, it could become “a civic-tech cash grab.”

Read article

AI and Labor

Bossware Comes to California: The Impact of New Management Technologies on Workers and the Case for Policy Regulation

Center on Policy Initiatives, East Bay Alliance for a Sustainable Economy, LA Alliance for a New Economy, Working Partnerships USA, PowerSwitch Action, UC Berkeley Labor Center, and UCLA Labor Center on October 8, 2026 in UC Berkeley Labor Center

A seven-organization research collaboration documents how “bossware” — electronic monitoring combined with algorithmic management — is already reshaping California workplaces. The tools include activity tracking, biometric timekeeping, productivity software, always-on cameras, and AI-generated emotion detection, deployed largely without disclosure to workers (41% of workers surveyed said nobody told them). The report makes the case for state regulation of workplace AI before it hardens.

Read article

AI and Labor

Working People Need Standing to Shape the AI Economy

Amanda Ballantyne on October 13, 2026 in New America (Workers & Technology)

Senior Fellow Amanda Ballantyne argues worker protections during the AI transition require more than shock absorbers (unemployment insurance, retraining) and guardrails (surveillance limits, no-robot-boss rules). The missing third category is “constitutive rights” — formal standing for workers and communities to shape AI itself in the venues where it’s governed: federal research funding, public procurement (OneGov, GSA deals), industrial policy conditions (CHIPS), market structure, and data governance. Drawing on John Kenneth Galbraith’s “countervailing power” and Daron Acemoglu and Simon Johnson on technological change, Ballantyne argues parity, not input, is what matters — and that bodies like the National AI Research Resource and CHIPS Industrial Advisory Committee need worker and community seats, not consultation rights.

Read article

AI and Elections

Anthropic’s Claude Accessed Georgia Voter Data Without Authorization

Maria Curi and Alex Isenstadt on October 25, 2026 in Axios

Anthropic disclosed that while staff were using Claude internally on a statistics project, Georgia voter data was accessed without authorization. The data was aggregate vote counts, which the state sells for a fee. Unprompted, the model requested a token to query a paywalled dashboard, and the fee was bypassed. The incident is the latest in a growing pattern of agentic deployments operating without authorization.

Read article

▲