U.S. Government Accountability Office Urges Facial Recognition Regulation

Executive *******

****.*. ********** ************** ****** (***)****** ********** **** *********** ****** ** open *** ***********, ******* ** ******** uses, ********** ** **** ******* ********* collected ****, ***** **** **** *** most ******** *** ***** ****** **********, and ******** *** **** *** **** protection. **** ****** ********* ******'*************** **** ******** ********** ******** ********** laws ** ******* *** **********. **** first **** **** ************** * ***** ago ** ** ****** *** **** recognition *** ******* **** **** ***** fire ********, ** *********** ********** ** more ******.

What ** ***

****.*. ********** ************** ******** * *********** **** ** *** U.S. ******** **** ***** ******** ******* issues. *** ** ******** ** ** the************* ********. **** *** ***** ****** **** ensuring ******** ********* *** ***** ***********.

Face *********** *** *****

*** ***** ********** **** *********** *** has ******** ***** ****, ****** ** cheaper *** ****** ********** **** **** providers. ** ********, *** ***** **** expanded. *** ***** **** ****** ** 2015, **** *********** ** ***** **** or ******** *** ******* **********, ********** tracking, *** ***** ******* *******.

**** ******* **** ********* ** ****** by ************ ** ********** *** ********* costs.

*******,**** *** ***** **** **** *********** use ** ********* *** ************ ******** *** ** high ******, **** ** ******, ******** over *******, *** ******** **** **********, such ** **** *** ** ************ here.

GAO **** **** ****** ******** *****

***** *** ********** *** ****** *** growing, **** ****** ****** ****** ******** *****, ** **** ******* *****:

**did *** ************* ****** the global facial recognition market revenues or forecasted revenue estimates. However, we did review information from several market research firms—Allied ****** ********, ******** *** *******, *****************, ******* ********, **********, ****** ******** ******, *** ******* ****** ********— and found that the estimates fell within consistent ranges, with only one outlier. [emphasis added]

**** ** * ******* ******* ** have **** ** ***** ************ *** now *** ** **********. **** *** data ***** **** *** ********** ** making ****** *********.

Existing ******* ***********

******** ** ******* **** *********** ******** legislation ******,******* * ** *** ******* ***** Commission (***) ****** ***** ** **** ***********, **** GAO. *** ************* ******** **** ***** ****** ******* ********* that *** ****** ** ********* ********* regarding ********. *** ******** *** *** can *** **** ** ** ***** companies *** ******** ** ******** ********* regarding **** ***********.

*** ***** **** *** *** ******** ****** ******* ************ *** ***************** ** *****' ******* involving ****** ***********. *** *********:

*** *** *** ****** ***** *** a**** *** ****** *********** **** ********* report. ** **, *** *** ******:

*. ******* ** ******: ********* ****** build ** ******* ** ***** ***** of ******* ***********.

*. ********** ******** ******: *** ********* that *** *** ********** **** *** context ** * *********** ** * consumer’s ************ **** * ********, ********* should ******* ********* **** ******* ** a ******** **** *** *******.

*. ************: ********* ****** **** *********** collection *** *** ********* ***********.

************, ************ *********** ******* ****** ********** facial ***********:

*****, **** ****** ****** * ********’* affirmative ******* ******* ****** ***** * consumer’s ***** ** *** ********* **** derived **** **** ***** ** * materially ********* ****** **** **** *********** when **** ********* *** ****. ******, companies ****** *** *** ****** *********** to ******** ********* ****** ** * consumer ** ******* *** ***** *** otherwise ******** *** ** ***, ******* obtaining *** ********’* *********** ******* *******

*******, ** ** ***** ************ ****** recognition, **** ****** *******.

*** *******, *** *** *********** **** demographic ********* ** ******* ***** ****** before ********* *** *******:

*** *******, ********* ***** ******* ***** capable ** *********** ********* – ***** often **** ** ********* **** ******* signs **** ** *** ******* ******* – ****** ******* ***** ****** ** consumers **** *** ************ *** ** use, ****** ********* **** **** ******* with *** *****.

** *** ***** ****, **** **********, this *** **** * **************, *** law:

*******, ** *** ****** *** *********** best ********* ** ****** ******** ***** requirements, **** *** *** ******** ** serve ** * ******** *** *** enforcement ******* ** *********** ***** **** currently ******** ** *** ***

State ****

*** *** ******* * ****** **** laws ********** ** **** ***********; **********, Illinois, **********, *** *****. *** **** explained **** ** *** ** ****** to ****** **** ******** **** ** standards, ** ********* *** ****** ** comply **** *** **** ********* ********* everywhere.

• *** ******** ********* *********** ******* ********* ************ ** ********* **** ********** and ***. ******** ******** ****** *** consent ** *******, ****, *** *** of **** ****.*** ***** **** ******* ****** ******** were ***** **** **** ******* **** they ******** **** ***.
• *** ********** ********* ******* *********** ****** *** ******* ** ********* data ********** ** **** ** * mechanism ** ******* ****** *** ** this ****.
• *** ***** ******* ** *** ******* or *** ** ********* ******************* *** ********** ** ********* **** without ****** *** ******* *** ****** its ***.
• *** ********** ******** ******* *********** ********* ** ******** *** **** they ******* (********* *********), ******* *** the **** *** **** ****, *** allow ********* ** ***-*** ** **********.

**** *** **** ****** *********** ********* opt-out ** ********* **** ******** ** Illinois, ** ** *******.

Existing **** *** ******

*******, *** *** *** ******* ***** existing **** ******** ** ******* ********* as *** **** ** *********, *******:

*** ********** *** ******** ********** **** a **** ****** (***-**-***) **** ******** ******** ************* *** consumer ******* ********* ** ******* ******* in ********** *** *** ***********.

Principles *** **** ***********

*** ****** **** *********** ****** ** open *** ***********, ******* ** ******** uses, ********** ** **** ******* ********* collected ****, ***** **** **** *** most ******** *** ***** ****** **********, and ******** *** **** *** **** protection. *** ********** **** **** ********** should ** *******, **** ******* ********* and *******. **** ****-******* ******** **** should ** ****. **** **** ****** be **** *** * ******** ******* stated *******. *** *** ***** ****** be *******, *** *** ******** ******* consent. **** ***** ****** ** ********** for **** ********** *** *** ********. Also, **** ********** **** *** ******** across ************ ****** ** ****.

****

*** ******** ******* **** *********** ***** ****** ******** * ** *********** *********** *** noted **** ****** ** **** *********** but ***** ** ** ********* *** the ***** ** ****. **** ***** that ***** **** ********** **** ******** and ********* ********* ******* *** ****** others ********* **** *********** ** ********* groups. *** ********* **** ***** ** no ********* *** *** ***** ** this ****. *** ********* **** ******** data ** ***** ****, *** ***** cause **** ** **********. **** **** that ***** ******* ** ** ** “other-race ******” ***** ********** ********* ***** on ****** *** **** *************** ********* than *** ********. *** **** ***** that ***** ******* ***** **** ******* effects ** ******** ********* ** * subject's ****.

Recommendations *** ********** ****

*** *********** **** ******* ******** **** are ****, ********** *** ******** ** avoid ****, ****-******* ****** *** ****, and ********** *** *********. **** ***** that ***** ******* ****, ** ******** in * *** **** ******* ******** sets *** *******, ***** ************ ****. *** ********* **** ********** can ** ******** ** ********* ****. Instead ** ************ ****** *****, ********* can **** ********** ****** ******* ***** across ********* ************.

*** ******:

********** ***** ****** ********** ** ******* equal ***** ***** ******* *********** ******

*** **** *********** **** ***** ********* were ******* **. **** ***** **** algorithms ****** ** **** *********** *** monitored. **** ******* ***-**** ******** ***** help ******** ****. *** ********* ********* modification ******* **** ******** *** ******** concern ********** *** ******** ** *********** bias.

***, ***** *************** *** ********** **** may ** ***********. ** ***** ** hard ** *** * ******* ******** set ******* ********* *******. ********** *** be ****** ** ****** ******* ********* data ** **** **** ** *** diverse. ****, ******** ** ******** **** through ******** *** ***** ****-*******. ** algorithm ***** *** ** ***** ****** on ***** *** (** ** *******) who *** *** ** *** ******** set, ******* ** ** ********* *** the ********** ***** *** ** *** training ****. ****, ******* ***** ***** rates *** *** **** ******** ** threshold ************. ** **** **** ********** less ******** ****-***, ***** *** ******** is * ******** *******.

IPVM ***************

***** *** **********'* ********** *** ****, the ********** **** **** ** ******* in * ****** ** *****.

(*) ******* ****-******* ******** ********** *** used.****'* **** *********** ******** ********* *** ** *** ******* submitted ********** *** ********** ********. **** means **** **** **** *********** *********, ranging **** ****** *** ******** ** most ********** ***** ************ *********, *** missing **** *** *****. *** **** the ***** ** *** *********** ********* the ********** * ******* *****. ****,*********** **** *********** ******* *** ****** available,****** ******** ******* *** ********** ****. The ********** ***** **** ** ******* some **** ** ******* ** ****** to ****** *******, ** **** ** for **** ***** ******** ******* **** medical ******** ** *********** ** ****, etc.

(*) ***-******* ****** *** * ****** part ** ************. ****** ********* *** to *** ***** ****** ** *********, but **** ** ******* *** ***********. The ********** ***** ****** **** ******* low-quality ****** ** ********* *** ********* improved *********** ** ****** ****-***** ***** surveillance *** *****.

(*) ********** ** *********** ***** ******** some **** ** ********* *** ******. Facial *********** ********, ******, ****** ************* determine ** * **** ** ***** it *** ********* ** ******* ********. The ********** ***** ******* ****** *********** users ** ******** **** ******* **** have **** *** ** ** *** public ** ***** ******* **** **** of ******** **** ********** *******. **** does *** ******* **** *** ******** would ** **** ** ****-******** **** as ******* **** *** **** ********** to **** ******** *** *** **** to **** ** ******** ******** ***.