Falls Church allows Copilot Chat but does not track AI use by staff

In short: Cities are trying to set rules for workplace AI, and Falls Church, Virginia is drawing attention for allowing one tool but not tracking how staff use it.

What's going on

Falls Church, Virginia says city staff may use Copilot Chat through their work accounts. Copilot Chat is a chatbot that can help write and summarize text, like an assistant that answers questions in a chat window. The city policy says employees cannot enter personally identifying city data into the tool.

The city also blocks consumer AI websites on city networks and devices. That means staff cannot access public chatbots from city computers, similar to how some workplaces block social media sites. These steps are meant to lower the risk of sensitive information being copied into the wrong place.

A key criticism in the discussion is that Falls Church does not log or audit AI use by employees. In plain terms, it does not keep a record of what people type into approved AI tools, and it does not monitor AI use on personal devices. The article suggests the city should log AI use, assign someone to review those logs, and name an owner for handling AI-related incidents.

The same debate included an example involving an AI “agent,” meaning an AI tool that can take several steps on its own, like a helper that can click around different systems. The example described an Anthropic Claude agent that, during a statistics project, reportedly asked for an access token (a special digital key) for a Georgia Secretary of State dashboard and then retrieved aggregate vote counts.

What to watch

Other governments are considering stronger safeguards. New York City lawmakers have discussed ideas like outside testing, protections for whistleblowers, and formal incident reporting, which is similar to requiring a written report when something goes wrong.

Source: NYTimes