Global declarations on AI governance abound—but the real test lies in implementation, much of which is unfolding in cities. Yet local initiatives are rarely monitored or shared across jurisdictions. The AI Localism Repository aims to bridge that gap by spotlighting governance mechanisms developed at the city level.

In this post, we highlight five recent additions to the repository that exemplify the diversity and innovation of AI Localism. These examples—spanning Reykjavík to Manchester—show how local governments are using distinct mechanisms such as risk-tiered staff guidelines, bright-line safeguards in high-risk domains, procurement nudges, AI-assisted public deliberation, and community literacy roadshows to build more responsive, inclusive, and future-proof approaches to AI.

Most Recent Additions

• Principles and Rights

The City of Reykjavik created its AI Framework for use of AI tools by city employees based on the degree of care with which city employees should exercise. Following a national data security framework, tools are classified as open use, protected use, highly protected use, or prohibited use. Open use tools are non-proprietary tools which can use open data inputs. Protected use tools are widespread AI tools approved by the city for internal operations and must undergo additional data governance processes. Highly protected tools must be used under strict supervision and only in tailored circumstances.

• Laws and Policies

Illinois Wellness and Oversight for Psychological Resources Act: llinois has banned the use of artificial intelligence to provide mental health services. In response to concerns over the harms of hallucinated or error-based health-related AI recommendations, the Act bans the use of AI in the direct provision of therapeutic services, decisions, or treatment plans.

• Procurement

Detroit RFP AI Clause: Detroit is integrating an AI invitation statement into all RFPs. They publish 300 RFPs per year and will now accompany each RFP with a statement saying that, “Contractors are encouraged to employ innovative approaches, including ethically and responsibly leveraging artificial intelligence and advanced technologies, to enhance goods delivery, services, and operational performance.” Use of AI will not impact RFP scoring.

• Engagement

Washington DC’s AI Public Listening Session: Washington DC’s Advisory Group of AI Values Alignment holds public listening sessions to establish values for the use of AI, and, for the first time on July 15, 2025, used an artificial intelligence tool to gather feedback during the listening session. If deemed successful, the tool, deliberation.io, will be rolled out for use in other agencies to “support large-scale open dialogue between residents and help the district gather feedback." 

• Literacy

Manchester (UK) People’s Panel for AI: Manchester is deploying a series of events to decrease the digital divide. The first stage was a series of roadshows where local university researchers introduced residents to AI basics before discussing the ethics of use and application in everyday life (i.e. student use of AI in essays, trustworthiness in social care)

Key Themes and Takeaways

1) Risk-tiered guidance beats blanket rules.
A layered approach gives frontline staff practical judgment without stalling useful experimentation. By classifying tools and uses according to data sensitivity and potential impact, cities can spell out what’s open by default, what requires extra governance steps, and what is off-limits. This shifts daily decisions from ad-hoc caution or unchecked enthusiasm to a predictable pathway where safeguards scale with risk, reducing ambiguity while keeping room for responsible innovation.

2) Draw bright lines where harms are acute.
Some domains—like mental health—carry stakes that are too high for probabilistic guardrails alone. Clear prohibitions set a protective boundary that prevents the quiet normalization of risky practices while still allowing supportive or administrative technologies. The clarity also helps vendors and agencies align expectations early, lowering compliance friction and avoiding harm-containing afterthoughts.

3) Procurement is a low-friction lever.
Adding a simple AI invitation statement to RFPs can surface better ideas without tilting competitions toward hype. It signals that the city is open to responsible AI but won’t reward it merely as a buzzword. Over time, this gentle nudge helps build a market of proposals that are concrete, testable, and auditable, while preserving fairness in scoring and reducing pressure to purchase solutions that are not fit for purpose.

4) Pair public engagement with the right infrastructure.
Listening at city scale requires more than town halls; it needs tools that structure input, widen participation, and make trade-offs visible. Piloting AI-assisted deliberation allows agencies to gather richer feedback and compare options transparently. The point is not to automate consent, but to make engagement repeatable, auditable, and easier to extend across departments once it proves inclusive and useful.

5) Make literacy local and lived.
People learn and deliberate most effectively when examples mirror their actual choices—how students write, how social care assesses need, how residents check information. Grounding literacy programs in familiar scenarios builds critical judgement rather than rote enthusiasm for tools. It equips communities to recognize when AI adds value, when it distracts, and when refusing or limiting use is the wiser path.

6) Share and iterate across jurisdictions.
Local efforts often disappear into internal drives or isolated pilots, forcing others to start from scratch. Publishing policies, templates, evaluations, and post-mortems turns each initiative into a shared asset that other cities can adapt. This culture of open iteration shortens learning cycles, surfaces common pitfalls early, and steadily raises the baseline for responsible, future-proof AI use in the public sector.

These snapshots show how cities of all sizes are setting and refining governance measures for the responsible deployment of artificial intelligence. For a deeper explanation of AI Localism, read our report AI Localism in Practice: Examining How Cities Govern AI. Do you know of an AI Localism measure that should be added to the repository? Use the form on our website to share examples of AI Localism or to express your interest in collaborating with us on the AI Localism program.