When Los Angeles Unified School District launched a districtwide AI chatbot nicknamed “Ed” in March, officials boasted that it represented a revolutionary new tool that was only possible thanks to generative AI — a personal assistant that could point each student to tailored resources and assignments and playfully nudge and encourage them to keep going.

But last month, just a few months after the fanfare of the public launch event, the district abruptly shut down its Ed chatbot, after the company it contracted to build the system, AllHere Education, suddenly furloughed most of its staff citing financial difficulties. The company had raised more than $12 million in venture capital, and its five-year contract with the LA district was for about $6 million over five years, about half of which the company had already been paid.

It’s not yet clear what happened: LAUSD officials declined interview requests from EdSurge, and officials from AllHere did not respond to requests for comment about the company’s future. A statement issued by the school district said “several educational technology companies are interested in acquiring” AllHere to continue its work, though nothing concrete has been announced.

A tech leader for the school district, which is the nation’s second-largest, told the Los Angeles Times that some information in the Ed system is still available to students and families, just not in chatbot form. But it was the chatbot that was touted as the key innovation — which relied on human moderators at AllHere to monitor some of the chatbot’s output who are no longer actively working on the project.

Some edtech experts contacted by EdSurge say that the implosion of the cutting-edge AI tool offers lessons for other schools and colleges working to make use of generative AI. Most of those lessons, they say, center on a factor that is more difficult than many people realize: the challenges of corralling and safeguarding data.

An Ambitious Attempt to Link Systems

When leaders from AllHere gave EdSurge a demo of the Ed chatbot in March, back when the company seemed thriving and had recently been named to a Time magazine list of the “World’s Top Edtech Companies of 2024,” company leaders were most proud of how the chatbot cut across dozens of tech tools that the school system uses.

“The first job of Ed was, how do you create one unified learning space that brings together all the digital tools, and that eliminates the high number of clicks that otherwise the student would need to navigate through them all?” the company’s then-CEO, Joanna Smith-Griffin, said at the time. (The LAUSD statement said she is no longer with the company.)

Such data integration had not previously been a focus of the company, though. The company’s main expertise was making chatbots that were “designed to mimic real conversations, responding with empathy or humor depending on the student's needs in the moment on an individual level,” according to its website.

Michael Feldstein, a longtime edtech consultant, said that from the first time he heard about the Ed chatbot, he saw the project as too ambitious for a small startup to tackle.

“In order to do the kind of work that they were promising, they needed to gather information about students from many IT systems,” he said. “This is the well-known hard part of edtech.”

Feldstein guesses that to make a chatbot that could seamlessly take data from nearly every critical learning resource at a school, as announced at the splashy press conference in March, it could take 10 times the amount AllHere was being paid.

“There’s no evidence that they had experience as system integrators,” he said of AllHere. “It’s not clear that they had the expertise.”

In fact, a former engineer from AllHere reportedly sent emails to leaders in the school district warning that the company was not handling student data according to best practices of privacy protection, according to an article in The 74, the publication that first reported the implosion of AllHere. The official, Chris Whiteley, reportedly told state and district officials that the way the Ed chatbot handled student records put the data at risk of getting hacked. (The school district’s statement defends its privacy practices, saying that: “Throughout the development of the Ed platform, Los Angeles Unified has closely reviewed the platform to ensure compliance with applicable privacy laws and regulations, as well as Los Angeles Unified’s own data security and privacy policies, and AllHere is contractually obligated to do the same.”)

LAUSD’s data systems have recently faced breaches that appear unrelated to the Ed chatbot project. Last month hackers claimed to be selling troves of millions of records from LAUSD on the dark web for $1,000. And a data breach of a data warehouse provider used by LAUSD, Snowflake, claims to have snatched records of millions of students, including from the district. A more recent breach of Snowflake may have affected LAUSD or other tech companies it works with as well.

“LAUSD maintains an enormous amount of sensitive data. A breach of an integrated data system of LAUSD could affect a staggering number of individuals,” said Doug Levin, co-founder and national director of the K12 Security Information eXchange, in an email interview. He said he is waiting for the district to share more information about what happened. “I am mostly interested in understanding whether any of LAUSD’s edtech vendors were breached and — if so — if other customers of those vendors are at risk,” he said. “This would make it a national issue.”

Meanwhile, what happens to all the student data in the Ed chatbot?

According to the statement released by LAUSD: “Any student data belonging to the District and residing in the Ed platform will continue to be subject to the same privacy and data security protections, regardless of what happens to AllHere as a company.”

A copy of the contract between AllHere and LAUSD, obtained by EdSurge under a public records request, does indicate that all data from the project “will remain the exclusive property of LAUSD.” And the contract contains a provision stating that AllHere “shall delete a student’s covered information upon request of the district.”

Rob Nelson, executive director for academic technology and planning at the University of Pennsylvania, said the situation does create fresh risks, though.

“Are they taking appropriate technical steps to make sure that data is secure and there won’t be a breach or something intentional by an employee?” Nelson wondered.

Lessons Learned

James Wiley, a vice president at the education market research firm ListEdTech, said he would have advised AllHere to seek a partner with experience wrangling and managing data.

When he saw a copy of the contract between the school district and AllHere, he said his reaction was, “Why did you sign up for this?,” adding that “some of the data you would need to do this chatbot isn’t even called out in the contract.”

Wiley said that school officials may not have understood how hard it was to do the kind of data integration they were asking for. “I think a lot of times schools and colleges don’t understand how complex their data structure is,” he added. “And you’re assuming a vendor is going to come in and say, ‘It’s here and here.’” But he said it is never that simple.

“Building the Holy Grail of a data-informed, personalized achievement tool is a big job,” he added. “It’s a noble cause, but you have to realize what you have to do to get there.”

For him, the biggest lesson for other schools and colleges is to take a hard look at their data systems before launching a big AI project.

“It’s a cautionary tale,” he concluded. “AI is not going to be a silver bullet here. You’re still going to have to get your house in order before you bring AI in.”

To Nelson, of the University of Pennsylvania, the larger lesson in this unfolding saga is that it’s too soon in the development of generative AI tools to scale up one idea to a whole school district or college campus.

Instead of one multimillion-dollar bet, he said, “let’s invest $10,000 in five projects that are teacher-based, and then listen to what the teachers have to say about it and learn what these tools are going to do well.”