Fresno city manager says government is 'captive' to AI vendors

City leader warns of vendor lock-in as AI costs remain opaque

Fresno City Manager Georgeanne White told reporters the city is "captive" to private technology vendors whose software and AI tools have become essential to municipal operations—and she doesn't see a way out.

The blunt assessment came during a June 2026 budget hearing when Councilmember Miguel Arias pressed for details on artificial intelligence spending. Arias warned that AI pricing models resemble predatory schemes: low initial costs followed by steep increases once governments are dependent.

White acknowledged the concern but framed vendor dependence as unavoidable. She compared today's AI adoption to the 1990s transition from WordPerfect to Microsoft Word—individual resistance eventually gives way to industry-wide standards. "You're either going to have technology or you're not," White said. "We are captive—we're absolutely captive."

A city spokesperson told Fresnoland that actual costs for AI programs, including usage-based Claude AI tokens, are not readily available.

No formal policy governs AI use

Fresno employees have used AI tools for work purposes for years without formal guardrails. The city only blocked OpenAI's ChatGPT after the company contacted officials demanding either an enterprise subscription or cessation of use. Employees were redirected to Microsoft CoPilot instead.

White said an informal policy exists and will eventually be formalized, but acknowledged recent problems. An outside law firm representing Fresno was sanctioned for citing invalid legal precedents likely generated by AI.

A city memo obtained by Fresnoland lists two dozen AI-enabled programs approved for employee use. Projected spending for fiscal 2027 includes approximately $25,000 each for Madison AI (which drafts staff reports), Salesforce AI features, and Wordly AI (providing live translation at council meetings). However, projections for Claude AI usage tokens were not provided.

Fresno police use AI built into Axon body cameras to draft reports. While the AI features carry no separate charge, the city's five-year Axon contract totals $18.3 million—rising from $1.2 million in year one to $5.8 million in year five.

Why it matters

Fresno's experience illustrates a broader challenge facing governments: once AI tools are embedded in core workflows, switching costs and vendor dependencies make it difficult to reverse course or negotiate better terms. Without clear policies on acceptable use, data privacy, and accuracy verification, municipalities risk compounding technical debt with governance gaps. The lack of transparent cost tracking also makes it nearly impossible for elected officials or taxpayers to evaluate whether AI investments deliver value—a problem that will intensify as more governments adopt these tools.

Researchers warn of risks

David Krueger, an AI safety researcher at the University of Montreal who founded the nonprofit Evitable, told Fresnoland that governments should establish AI policies before deployment, not after. He recommends maintaining records of all AI applications, blocking tools that leak private data or train on user inputs, and ensuring employees understand AI hallucinations and biases.

A 2026 paper from Wharton researchers found that AI use can lead to "cognitive surrender," where users adopt AI-generated answers rather than constructing their own reasoning—even when outputs are flawed.

Some major corporations have already pulled back. In May 2026, Uber's COO said the company exhausted its annual AI budget by April with little customer benefit to show for it. That same month, Starbucks canceled an AI inventory system after it repeatedly miscounted items across thousands of stores.

These details were first reported by Fresnoland's Omar S. Rashad.