Some US unemployment applicants say facial recognition models used by ID verification service ID.me are failing to identify them, locking them out of benefits
Todd Feathers / VICE :
Context & Ripple Effects
ID.me's facial recognition check has become a de facto gate on US unemployment benefits, and applicants the models cannot identify are cut off from income they are legally owed. The failure mode echoes an earlier automation disaster in this exact domain, when Michigan's flawed unemployment system falsely charged thousands with fraud — states outsourcing eligibility decisions to software with thin accountability.
The stakes grew after this report: ID.me went on to win verification of IRS.gov logins while claiming a $400B theft in pandemic unemployment payments, and a later House investigation concluded those fraud claims were baseless marketing designed to increase demand. This article is the first crack in that story.
First-order effects
- Unemployment applicants who fail ID.me's face match are denied benefits through no fault of their own, with the burden of appeal falling entirely on them while states treat the vendor's verdict as authoritative.
- State workforce agencies relying on ID.me absorb the complaints and reputational damage of a verification system they did not build and cannot easily audit.
Second-order effects
- Other government-facing identity vendors now compete on accuracy and appeal processes rather than just price, since a single high-profile failure rate becomes procurement ammunition for rivals.
- As ID.me expanded toward verifying IRS.gov logins, the same model failures would scale from state benefit offices to federal tax accounts, raising the cost of every false negative.
Third-order effects
- The pattern — automated gatekeeping followed by inflated claims of the fraud it prevents, as the House probe later found with ID.me's baseless COVID-19 fraud claims — pushes regulators toward mandatory human review and error-rate disclosure for biometric systems used in benefits determinations.
- Government agencies face structural pressure to keep algorithmic verification reversible: any system that locks people out of money must ship with a non-biometric fallback path, or procurement itself becomes the liability.
The trend: Biometric identity checks are becoming default infrastructure for government benefits, and each documented failure converts accuracy metrics into a public-accountability issue for the vendors states hire.