In recent months, Elon Musk’s DOGE has attempted to hollow out America’s federal bureaucracy. After firing droves of workers and attempting to downsize prominent agencies, the billionaire-backed effort is attempting to create a new governance model that prioritizes automation. On Friday, Wired reported that, in a flailing attempt to modernize the agency, a new ChatGPT-style bot was integrated into agency staffers’ workflows.

The Agency Support Companion is supposed to “assist employees with everyday tasks and enhance productivity,” an internal email viewed by the magazine reads. However, the chatbot does not seem to work very well. “Honestly, no one has really been talking about it at all,” a source who works at the agency told Wired.

The app’s launch was reportedly accompanied by a hilariously terrible training video (seen here) that involved a poorly animated, four-fingered woman. The video, which was supposed to explain to staffers how to use the app, neglected to tell them a very critical piece of information: that they should refrain from uploading sensitive personal information to the program. This oversight forced the agency to send out an apologetic email to staffers that highlighted the missing context: “Our apologies for the oversight in our training video,” it read.

“I’m not sure most of my coworkers even watched the training video,” the SSA source told Wired. “I played around with the chatbot a bit, and several of the responses I received from it were incredibly vague and/or inaccurate.” They added: “You could hear my coworkers making fun of the graphics. Nobody I know is [using it]. It’s so clumsy and bad.”

If Musk’s plan is to automate the SSA, there is plenty of evidence (aside from the terrible new app) that suggests it’s a bad idea. Indeed, a similar attempt at automating social services in Brazil shows why over-reliance on algorithms to operate welfare programs may result in worse outcomes for everybody.

Rest of the World reports that an attempt by the Brazilian government to reduce bureaucracy by replacing officials with algorithms hasn’t always produced the best results. Brazil has an app, Meu INSS, that was developed by a state-owned company, Dataprev, and is designed to handle social security claims. The app, which was launched in 2018, uses computer vision and natural language processing to analyze documents submitted to the government by claimants. Unfortunately, the app has a habit of rejecting legitimate claims based on minor errors. Those automated decisions can kick off lengthy legal battles that take many months to resolve.

The outlet documents the experience of one former sugarcane worker, 55-year-old Josélia de Brito, who filed for her retirement benefits through the app but was mistaken for a man by the automated system and had her benefits denied. “I have all the documents proving my health condition, proving everything, and [the benefit] still gets denied. It’s a humiliation,” De Brito told the outlet.

The nation’s rural farmworkers have struggled with the increasingly digital face of social services, the outlet notes. “People out here cannot [even] work with Gmail, Facebook, Instagram,” Francisco Santana, president of the Union for Rural Workers at Barra do Corda, told Rest of World. “Processes are [getting] more and more automated, and society wasn’t made ready for it, especially further away, in the outskirts, for people that live in rural areas.”

Other nations’ struggles with automation are a potential warning for the U.S., whose social services are currently being “modernized” by technocratic minions deployed by the Trump administration. Musk’s DOGE has attempted to gut agencies across the federal bureaucracy, including the SSA. In place of full staff, DOGE has sought to institute an “AI-first” strategy that would cut the federal workforce in half and replace human workers with software. “Everything that can be machine-automated will be. And the technocrats will replace the bureaucrats,” a person knowledgeable about DOGE’s activities recently told the Washington Post. Far from resulting in greater efficiency, however, DOGE’s changes have so far helped to spur dysfunction and chaos at many agencies, including Social Security.

Indeed, a recent incident at the SSA, in which DOGE workers reportedly marked “countless” living benefit recipients as “dead” and cut off their benefits, is just one example.  “[DOGE staffers] went into the system and they killed off people,” Rennie Glasgow, a longtime claims technical analyst at the agency, told The Daily Beast. “About 4 million people, they marked them as dead. But they’re not sure if those people were supposed to be marked as dead, so they’re sending us an email saying, ‘If these people come into the office with their identification, you can reinstate them,’” Glasgow told the outlet.

“We have to go through this long process to resurrect them, to get them back as alive, which can take about three to four days,” Glasgow added.

DOGE also recently announced that it would attempt to rewrite the entirety of the SSA’s codebase in a matter of months. Sources have suggested that, at the breakneck speed at which they plan to work, DOGE would almost certainly have to rely on AI to do this. While writing code with AI has become a more common practice, it’s also one that needs heavy supervision due to the degree to which the software makes mistakes. Given how many mistakes that DOGE’s human operators have made during their blitzkrieg of the government, it stands to reason that an automated rewrite of the SSA’s codebase could be a disaster.

Of course, some speculate that making mistakes is the point. Those critics believe DOGE isn’t attempting to “modernize” or improve the SSA at all, but is actually attempting to destabilize and destroy the agency so that it can be privatized. If that’s the case, then everything is going to plan.