IBM facial recognition contract provokes debate on specifics, and ‘general purpose’ fear
A contract won by IBM last month to supply facial recognition technology to UK border security and law enforcement has prompted accusations that the company has abandoned prior commitments to limit its sale of the technology.
The contract with Home Office, originally listed as being for £65.6 million (US$84.4 million) but now corrected to £54.7 million (US$69 million), is to provide infrastructure and biometric matching for immigration controls, the issuance of passports and visas, and law enforcement investigations.
In 2020, IBM committed to cease selling “general purpose” facial recognition, and to avoid providing face biometrics for applications that involve “mass surveillance” or “racial profiling.”
The use of facial recognition for one-to-many matching violates those commitments, according to some observers among media and rights advocacy organizations. The Verge quotes representatives of Black Lives Matter UK and Amnesty international that the contract is a reversal by the U.S. tech giant. The latter claims that “research across the globe is clear; there is no application of one-to-many facial recognition that is compatible with human rights law.”
One-to-many facial recognition is used by border and law enforcement agencies in dozens of countries around the world, including the U.S. and UK.
“IBM no longer offers general-purpose facial recognition and, consistent with our 2020 commitment, does not support the use of facial recognition for mass surveillance, racial profiling, or other human rights violations,” IBM spokesperson Imtiaz Mufti told The Verge.
Further, Mufti argues that because the platform is not capable of ingesting video, it does not support typical mass surveillance operations.
An IBM policy paper makes clear that the company has been advocating for export controls and other regulation of 1:N facial recognition software since 2020, and uses but does not specifically define the term “general purpose IBM facial recognition and analysis products.”
Doing so now runs the risk of appearing to move the goalposts. More likely, it would be widely ignored.