IRS Cancels Plans to Use ID.me Over Privacy and Accuracy Concerns
The IRS today ceased its plans to require citizens to use ID.me, a facial recognition service, to access their tax records following noted reconsideration from the Treasury department in recent weeks, now moving away from facial recognition altogether. The change comes after the CEO of ID.me, Blake Hall, initially stated that the company had no plans to use the “more complex and problematic” facial recognition software of one-to-many matching, which searches a large internal database for a match to a photo uploaded by the user, in favor of one-to-one matching, which matches a user’s uploaded photo to their official government photo. However, the company backtracked on this when it admitted to the use of one-to-many matching days later. One-to-many matching is controversial in the AI community, as it infamously has shown poorer accuracy for darker faces, as well as raised security risks over the storing of large amounts of biometric data, especially out of the hands of the government. While ID.me offers the chance for individuals to prove their identity over live video chat if the facial recognition service does not recognize them, there are fewer than 1000 agents assigned to this task for a service required by all Americans, and wait times have been reported to be hours long. The software at the core of the technology in use by ID.me, Amazon’s Rekognition, is no longer allowed to be sold to law enforcement agencies as a result of potential bias in the platform. Additionally, the general use of ID.me facial recognition technology for a service as basic as accessing tax records also puts Americans without smartphones or computer cameras at a disadvantage. Although ID.me is still widely in use by 10 federal agencies, including Social Security and Labor, 30 states, and 540 companies, by abandoning ID.me and its one-to-many matching, the Treasury Department and IRS took an important step in the direction of greater regulation of powerful, potentially dangerous facial recognition technology.