Two Contracts, No Public Vote: What LA and Baltimore Teach Governors

Two contracts, one pattern

In 2021, Los Angeles Unified School District signed a $6 million contract with a startup called AllHere to build "Ed", an AI chatbot meant to serve as a digital companion for 500,000 students and parents. The company had raised $12 million in venture funding and boasted major-district clients. By June 2024, AllHere had furloughed most of its staff. By July it had filed for Chapter 7 bankruptcy. Its founder, Joanna Smith-Griffin, was later charged with defrauding investors, accused of misrepresenting the company's revenue and customer base for years, according to reporting by The 74 and an indictment covered by the Associated Press. LAUSD's superintendent came under scrutiny over the deal.

In Baltimore, the school board approved a four-year, $5.5 million contract for Evolv, an AI weapons-detection system now used at entrances in 26 high schools. According to reporting from the Baltimore Sun, the contract was approved in February 2024 as one line in a nine-item funding vote, with no dedicated public discussion beforehand.

Neither case is really a story about AI failing on the merits. It is a story about how little scrutiny AI contracts got before money moved. That is the detail worth sitting with as English school leaders finalise their own procurement decisions this autumn.

Why this matters beyond the US

England's schools are not writing multi-million-pound cheques for chatbots, and the governance structures are different: academy trusts and maintained-school governing bodies both have statutory duties around financial oversight that a US district board vote does not map onto directly. But the underlying mechanism transfers. AI tools are pitched with confident demonstrations, a persuasive founder, and a promise of saved staff time or improved safety. Procurement decisions for software often move faster and with less scrutiny than decisions for, say, a building contract of similar value, because software doesn't feel like capital spending. Trustees and governors approving a subscription line in a wider budget paper can miss exactly what LAUSD and Baltimore missed: nobody independently tested whether the tool did what the pitch said, and nobody set out what evidence would count as proof it was working.

The failure in both cases wasn't a bad algorithm. It was a governance process that let a confident pitch substitute for evidence.

What US states are now doing about it

The response has been to write down the questions boards should have asked in the first place. The Southern Regional Education Board, a consortium of 16 US states, published an AI procurement and evaluation checklist for K-12 systems. Pennsylvania's Department of Education now instructs districts to establish, before signing: who controls the data entered into the tool, what third-party data-sharing the vendor does, whether data collection is limited to what's necessary, whether a human stays in the loop for any grading or discipline use, and how the district will assess whether the tool actually improves outcomes.

None of this is exotic. It is closer to due diligence than to AI policy. What is notable is that it took a bankruptcy and a fraud indictment on one coast, and a bundled vote on weapons-detection software on the other, before districts wrote it down formally.

What to do

For governors, trustees and business managers signing off AI spend this term, the transferable checklist is short:

• Separate AI contracts from routine budget lines. A five- or six-figure AI subscription should get its own agenda item and its own paper, not a row in a consolidated finance report.
• Ask what evidence exists before, not after. Request independent evidence the tool works for the stated purpose; a case study from the vendor is not independent evidence.
• Name a data owner. Know who controls pupil data entered into the tool, where it is processed, and what happens to it if the vendor collapses, as AllHere's did.
• Set a review date and a kill criterion. Decide in advance what result would trigger cancelling the contract, before the renewal conversation arrives under sunk-cost pressure.
• Let staff test it first. Both failed contracts involved tools rolled out to students and families without a meaningful adult pilot phase.

What to watch

Watch for whether England's own procurement frameworks, DfE guidance and the growing academy trust MAT-level policies, start requiring anything like Pennsylvania's checklist as a condition of spending. At the moment, the requirement to demonstrate evidence before buying exists in isolated state laws in the US and in England mainly through general public procurement rules, not AI-specific ones. The gap between a confident pitch and a working tool is where both LA and Baltimore's problems started, and it is a gap no policy has yet closed by default.

Sources: Route Fifty, Stateline, EdWeek, The 74 - AllHere bankruptcy, The 74 - federal fraud probe