FTC × IntelliVision: a 'zero bias, millions of faces' AI claim, measured against NIST

The problem

IntelliVision Technologies Corp., a San Jose company, sold an AI-based facial recognition product to OEMs, integrators and large end users. Its software was embedded in consumer products sold by its former parent Nice North America, LLC — the 2GIG Edge home security system and the Elan Intelligent Touch Panel — where, per the FTC’s complaint, “The software allows consumers to register their face and then scan their face to gain access to the system.”

From late 2018 through early 2024 the company’s website advertised, among other claims, “Zero gender or racial bias through model training with millions of faces from” datasets from around the world. In presentations to trade customers, “Respondent claimed its facial recognition technology has one of the highest accuracy rates on the market.”

What was built

An artificial-intelligence-based facial recognition and anti-spoofing (liveness) system. The FTC complaint records that “Respondent submitted its facial recognition algorithms to NIST for testing at various points in 2019, 2022, and 2023” — which is what makes this case unusual: the vendor’s marketing claims can be checked against a government laboratory’s published evaluations of the vendor’s own submitted algorithms.

The outcome

This is an honest-negative. The measured outcome is the gap the regulator found, and the order it imposed.

On training data. The complaint states: “Respondent also did not train its facial recognition software on millions of faces.” Instead, per the FTC’s December 2024 release, “the complaint alleges that IntelliVision did not train its facial recognition software on millions of faces, as it claimed, and instead trained its technology on images of approximately 100,000 unique individuals, and then used technology to create variants of those same images.”

On accuracy and bias. “The test results on NIST’s public website indicate that error rates for IntelliVision’s algorithms differed across different demographics, including region of birth and sex.” On false non-match rate, IntelliVision’s “algorithms were not among the top 100 best performing algorithms tested by NIST as of” December 19, 2023. And: “Respondent does not possess testing to support its claims that its facial recognition” technology has one of the highest accuracy rates on the market, can detect faces of all ethnicities without racial bias, or performs with zero gender or racial bias.

The consequence. Decision and Order, Docket No. C-4809 — “ISSUED: January 8, 2025”, published and announced on 13 January 2025. “After receiving no public comments, the Commission voted 5-0 to approve the final order.” The order “is final and effective upon the date of its publication on the Commission’s website (ftc.gov) as a final order. This Order will terminate 20” years from the date of its issuance. It bars misrepresentations about accuracy or efficacy, about comparative performance across genders, ethnicities and skin tones, and about spoof detection — and bars any such representation at all “unless Respondent possesses and relies upon competent and reliable testing that substantiates the representation at the time the representation is made.”

There is no fine. The order imposes no monetary relief. The $51,744 figure that circulates with this case is the FTC’s standard boilerplate maximum for a future violation of a final consent order, not a penalty assessed here.

Posture. This is a settlement. The Decision and Order records “statements by Respondent that it neither admits nor denies any of the allegations in the Complaint”. Every allegation-side figure above is the FTC’s, stated in its complaint.

As the Director of the FTC’s Bureau of Consumer Protection put it: “Companies shouldn’t be touting bias-free artificial intelligence systems unless they can back those claims up,” and “Those who develop and use AI systems are not exempt from basic deceptive advertising principles.”

Path to green

The subject of this story is the respondent, so there is no client-confirmation call that could earn a green badge for a delivery outcome. The verification step is documentary and must be re-run on the publication date:

1. Confirm from the FTC case page for Matter 232-3023 / Docket C-4809 that the order is still in force, unmodified, with no enforcement action filed against it.
2. Re-fetch the Decision and Order and the final complaint and confirm the SHA-1 digests still match the recorded Wayback CDX digests.
3. Ask the FTC’s Bureau of Consumer Protection whether the Part IV.A one-year sworn compliance report (due on or about 8 January 2026) was filed.

Until a human completes step 1 on the day of publication, this stays below green.