A Brooklyn Nonprofit Builds Open AI Tools for City Agencies

NEW YORK — In a shared workspace in Gowanus, a small nonprofit is building artificial-intelligence tools for city agencies with an unusual condition attached: everything they make is open for public inspection. The group's founders argue that as governments rush to adopt AI, they risk becoming dependent on opaque commercial systems whose inner workings neither officials nor the public can examine — a bargain the nonprofit was formed to resist.

The organization's projects are modest but concrete: a tool that helps a city office triage the flood of constituent complaints, a system that translates public notices into the many languages New Yorkers speak, a model that flags permit applications likely to need extra review. In each case, the group publishes its code and documentation, so that any agency — or any curious resident — can see exactly how a decision was reached. "Public software should be as inspectable as a public record," one co-founder said.

The push reflects a broader unease about the role of automated systems in government, where a flawed or biased model can quietly shape decisions about benefits, enforcement, or services. Advocates for algorithmic accountability have warned that proprietary systems make it nearly impossible to audit such decisions or contest them. The nonprofit positions its transparency as a direct answer to that concern. "If a machine helps decide something about your life, you should be able to ask it to show its work," the co-founder said.

City technology officials have expressed cautious interest, drawn by the prospect of tools they can adapt freely and are not locked into paying a single vendor for indefinitely. But adoption inside government is slow, tangled in procurement rules that were not written with open-source software in mind. The nonprofit's small size and grant-dependent funding also raise questions about whether it can support agencies over the long haul, a worry officials have voiced candidly.

Commercial AI vendors, for their part, argue that enterprises need the support, security guarantees, and accountability that come with a paid product, and that open-source alternatives can shift hidden costs onto overstretched agency staff. The nonprofit acknowledges the tension, saying its aim is not to replace every vendor but to give the public sector leverage and a baseline of transparency it currently lacks.

For the group's handful of engineers, many of whom left better-paying industry jobs, the work is partly ideological. "I spent years building things nobody outside the company could ever see," one engineer said. "Now I build things anybody can take apart. That feels like what public technology is supposed to be." The nonprofit said it hoped to pilot two more tools with city offices before the end of the year.