Issues With Face Recognition Software Used to Verify Unemployment Recipients Is Costing People Benefits

The facial recognition program used by 22 U.S. states to reduce unemployment fraud has been failing to correctly identify recipients, causing dozens of people to be denied benefits or have their applications put on hold, Motherboard reports.

Unemployment recipients have been sharing complaints online for months about the identity verification software ID.me, but the outrage boiled over this week in response to an Axios newsletter about the threat of unemployment fraud based on statistics provided by ID.me.

The service uses biometric data and official documents to verify users. However, several unemployment applicants told Motherboard that ID.me’s facial recognition models failed to correctly identify them, and they’ve had difficulty contacting ID.me’s support staff to fix the issue. Their applications were put on hold in the meantime, forcing should-be-beneficiaries to wait days or weeks until they could get in touch with an ID.me “trusted referee” to confirm their identity. Others who tried multiple times to pass ID.me’s facial recognition test found themselves locked out of the system.

ID.me CEO Blake Hall said in a statement to Motherboard and other outlets that the algorithms behind its Face Match technology have “99.9% efficacy.” In general, modern facial recognition tech has repeatedly proven to be less accurate for people of color. But Hall argued ID.me doesn’t suffer from this issue: A regression analysis the company performed found “no relationship between skin tone and Face Match failure on a 1:1 basis,” he said.

As for what could be causing these reported issues, Hall suggested user error may be to blame. “For example, if someone uploads a selfie that only shows half their face,” he told Motherboard. That could ostensibly cause the facial recognition software to fail to correctly identify them.

In his statement, Hall added that the company was not aware of “eligible individuals” who had been unable to verify their identity with its service. People who fail ID.me’s facial recognition test are not blocked, he continued, and can still prove their identity via video chat, which he claimed the wait time for is currently “less than five minutes and it has been consistently under 30 minutes all week.”

However, many people have taken to Twitter to say otherwise. Dozens complained that their emails, phone calls, and chat requests have gone unanswered for weeks or even months. Earlier this year, benefit recipients in Colorado who had no issues confirming their identity before the state’s labor department started using ID.me said they were suddenly rejected under the new system and went months without receiving payments. Similar stories made headlines in Florida, North Carolina, Pennsylvania, and elsewhere.

The rate of unemployment has soared amid the coronavirus pandemic, overwhelming many state offices and pushing their already antiquated computer systems beyond breaking point last year. As a result, some states reported a sharp rise in fraudulent claims last spring, and the Department of Labor said in February it had identified $5.4 billion of potentially fraudulent unemployment payments that went out between March and October of last year. That’s still a far cry from Hall’s estimate, who claims that as much as half of the unemployment payments in the U.S. during the pandemic, amounting to some $400 billion, were fraudulent claims.