Privacy Considerations for Choosing a Facial Recognition System

Though this is definitely one of the less inspiring aspects of evaluating a facial recognition vendor, it can save you an awful lot of time otherwise wasted in research.

A privacy policy not only reveals how a vendor handles the biometric information you submit to them and who they share it with both internally and externally, but also how diverse their organization is, their partners and affiliates, the different types of technologies they offer, and in general their overall understanding and assumptions about biometric privacy, which can vary depending on factors such as company and national culture.

The problem is that privacy policies are inherently boring to read. So, one good strategy is to start reading it in the middle, then pick random paragraphs north and south of the document. Sounds crazy, but if the alternative is you falling asleep, or just giving up after the first few paragraphs, this approach is more likely to help you find the most useful pieces of the document faster.

Important: If you view a vendors "privacy policy" and it turns out to be an End User License Agreement, or some other unrelated legal document, then the warning bells should start sounding in your mind about their overall understanding of the meaning of privacy. Yes, these types of vendors are out there ;)

Cloud-based facial recognition systems - ones that require an active connection to the Internet to function, are simply a bad idea, since many of them are amassing the photos and videos that are being submitted when you use the service, and these can then be unknowingly shared with, or sold off to third parties, including other companies and governments with more nefarious intentions. Additionally, the use of a centralized storage approach makes their services "high value targets" for cybercriminals and nation states.

A common response for vendors providing cloud-based services, who are questioned about the transfer of their customers data is "But, we have servers based in your country". While facial recognition vendors might claim to base their services and customer data storage in your organizations country of origin, in many cases these services merely act as a collection point for customer information, prior to transfer to the vendors home country.

As an example, Amazon S3 services are commonly exploited by some foreign companies in more geopolitically aggressive countries, as a data collection point prior to transferring data to their home country. This helps avoid being flagged by corporate firewalls, along with the resulting bad press, should their products attempt to transmit data directly back to their home country.

In a nutshell, by using a cloud-based service, you are giving the vendor the immutable biometric data of individuals, and it'll be difficult to get true visibility on what they are actually doing with this information over time.

The government of some countries, particularly ones that have a mindset of geopolitical domination, are far more aggressive in the way that they collect and use the biometric information of individuals. Know which country the company is based in. These vendors might also be working with their local government on facial recognition projects, which increases their motivation to exfiltrate facial profiles from your on-premise installation.

Additionally, the government of this country likely has eminent domain over the data being collected and amassed by the vendor, with the ability to legitimately seize the vendors information. Submitting facial profiles and other personally identifiable information to the vendor means that you are also potentially submitting it to their countries government.

To give you an idea of the breadth of facial recognition vendors out there, below is a breakdown by country, of the approximately 60 mostly vendors & a few universities that currently participate in the NIST Ongoing Face Recognition Vendor Test (FVRT). Keep in mind that there are additional entities developing facial recognition that don't currently participate in this testing.

• The Labeled Faces in the Wild (or LFW) sample set maintained by the Computer Science Department at the University of Massachusetts, Amherst, US


• The the color FERET database maintained by NIST


• The FICV datasets maintained by the Università di Bologna, Italy


• The Yale Face Database maintained by the Computer Science Department at Yale University, US


• The Microsoft Celebrity Dataset (MS-Celeb-1M) maintained by Microsoft, US


• Other sources of "In the Wild" facial profile repositories, and other evaluation guides can be found at the Face Recognition Homepage, maintained by Prof. Mislav Grgic, Ph.D., and Kresimir Delac, Ph.D. in Zagreb, Croatia.
  

On-premise facial recognition systems allow customers to build their own private data sets of facial profiles, and query them privately to protect the privacy of their customers.

If you evaluate or deploy an on-premise facial recognition system, this should be constantly monitored to understand whether the vendor transfers any of this private biometric information back to their network. This is because once installed and configured, an on-premise facial recognition system can be modified by the vendor over time to alter what they do with submitted facial profile information.

Ideally, no traffic should be allowed between the on-premise installation, and the Internet. The vendors installation should be able to operate entirely offline, with the ability to be updated via periodic updates downloaded from a separate Internet connected system.

Understand how secure the vendors on-premise installation is. This includes methods of authentication, how accounts, hardware and software are secured with firewalls, malicious code scanning, and how vendor's software updates are issued, tested and managed.

Ideally, no traffic should be allowed between the on-premise installation, and the Internet. The vendors installation should be able to operate entirely offline, with the ability to be updated via periodic updates downloaded from a separate Internet connected system.

If not already, your organization should develop an internal policy which includes how your customers personal information should be handled by employees with respect to privacy and the law, and the consequences when it is not. This can be included with any training that is facilitated as part of the deployment of your internal facial recognition systems.

An internal data handling policy helps to prevent the mishandling of personal information over time, through organizational structure and people changes.

An example of a policy template for law enforcement entities is provided by the US Department of Homeland Security

Over time, it's important to monitor your vendor's acquisition, merger, and investor status. A merger, acquisition, or change in investors of a company can change the nature of it's products and services. For instance their products might be neglected, put into maintenance mode, or modified to be more aggressive, or more neglectful when handling your customer's biometric data.

In the event that your vendor undergoes a major structural change, it pays to have an alternate facial recognition vendor in mind that can continue to regard the privacy of your customers information, or more actively maintain their facial recognition offering.