Google says it won't sell general-purpose facial recognition APIs before working through “important technology and policy questions”
More than 400 million people in the world have diabetes. A third of them have diabetic retinopathy, a complication that can cause permanent blindness.
Context & Ripple Effects
Google's 2018 pledge to hold back general-purpose facial recognition APIs sits at the center of a tension running through its later record. On one side is the restraint position: the company said it would work through "important technology and policy questions" before selling the capability broadly. On the other side are the practices that followed — a contractor, Randstad, allegedly using dubious tactics targeting people of color to collect facial scans for Pixel 4's face unlock, and an investigation finding Display & Video 360 letting advertisers target Americans' sensitive personal data despite Google's own rules against it.
First-order effects
- Google's cloud customers cannot buy a general-purpose facial recognition API from Google, so any face-matching use case has to be built as a bespoke internal project — like the Pixel 4 face unlock pipeline that relied on contracted data collection rather than a public product.
Second-order effects
- The gap between the API moratorium and actual conduct creates a credibility problem for Google: critics can point to the Randstad scanning allegations and the Display & Video 360 findings as evidence that internal rules, not just external sales, are where enforcement fails.
Third-order effects
- If voluntary "we'll wait until the policy questions are answered" stances keep coexisting with misuse allegations, the likely structural outcome is regulators treating biometric AI as too risky for self-governance — with Google's own missteps becoming case studies for mandatory oversight.
The trend: Google is expanding AI into high-stakes perception tasks — medical imaging, biometrics, ads targeting — while its stated safeguards lag behind what its systems actually enable.