Clearview Is So Toxic Even Other Surveillance Tech Purveyors Want Nothing To Do With It

from the bring-back-spitting-in-disgust-old-world-style dept

Outside of Clearview's CEO Hoan Ton-That, it's unclear who truly likes or admires the upstart facial recognition tech company. In the short time since its existence was uncovered, Clearview has managed to turn itself into Pariah-in-Chief of a surveillance industry full of pariahs.

Clearview hasn't endeared itself to the sources for its 10-billion image database, which are (in descending order) (1) any publicly-accessible website/social media platform, and (2) their users. The company has been sued (for violating state privacy laws) in the United States and politely asked to leave by Canada, which found Clearview's nonconsensual harvesting of personal info illegal.

It has subpoenaed activists demanding access to their (protected by the First Amendment) conversations with journalists. It has made claims about law enforcement efficacy that have been directly contradicted by the namechecked police departments. It has invited private companies, billionaire tech investors, police departments in the US, and government agencies around in the world to test drive the software by running searches on friends, family members, and whoever else potential customers might find interesting/useful to compile a (scraped) digital dossier on.

Clearview intends to swallow up all the web it can. Caroline Haskins' report for Business Insider (alt link here) catches Clearview's vice president of federal sales pretty much saying the only way to avoid being added to Clearview's database is to stop being online.

"People are constantly dumping their — it's just a constant," Clearview's Jones said during the roundtable event, referring to the steady stream of people posting their photos online, only for Clearview to scrape them.

The same report shows why Clearview's is so bad at PR. It has a PR team -- one that continually stresses the AI does not provide "matches," but rather "investigative leads." The difference between the terms is the extent of liability. If Clearview only generates leads, it cannot be blamed for false positives. If it says it delivers "matches," it possibly can be sued for wrongful arrests.

But CEO Hoan Ton-That can't stop using the terms interchangeably, which indicates he feels his software generates matches, thus inviting additional culpability.

"No one has done what we do at this scale with this accuracy," he said later in the conference, adding that anyone "is able to solve cases instantaneously if they get matches in the system."

The "accuracy" Ton-That claims no one can match is "98.6% accuracy per one million faces" -- an assertion Clearview's made for a couple of years now. Whether this claim is anywhere near 98.6% accurate remains to be seen. The AI has never been independently tested or audited.

We know how Clearview feels about itself. But how do the competitors in the market feel about it? Caroline Haskins spoke to other surveillance tech purveyors at the Connect:ID industry event and found other providers of facial recognition AI weren't thrilled Clearview was out there giving already-controversial tech an even worse reputation.

Several industry professionals openly expressed not liking or respecting the company. One government contractor said Clearview was "creepy." He told me he'd read about the company's extensive ties to the far right and was alarmed by that.

In one discussion, an attendee called the company "the worst facial-recognition company in the world."

Well, there's one thing surveillance tech purveyors and surveillance tech targets can agree on: Clearview took a bad thing, made it worse, and seems willing to trade paint with the internet at large to maintain its "Images in a facial recognition database [privately-owned]' lead.

While it's normal for competitors to criticize their rivals, the statements made here imply Clearview is bad for the facial recognition tech industry -- something that threatens these competitors' livelihoods. And that's not acceptable, even if the collective marketing of problematic AI to government agencies doesn't appear to bother them a bit.

Filed Under: facial recognition, surveillance
Companies: clearview

Australian Privacy Commissioner Says 7-Eleven Broke Privacy Laws By Scanning Customers' Faces At Survey Kiosks

from the small-privacy-breach-with-larger-repercussions dept

Of all the places to come across illegal facial recognition tech deployment, a convenience store chain is certainly one of the strangest. The tech wasn't deployed to stop shoplifting or keep unwanted people off the premises. Instead, somewhat ironically, it was deployed to help 7-Eleven convenience stores quantify how well it was doing in the customer service department.

Here's Campbell Kawn for ZDNet (via Slashdot):

In Australia, the country's information commissioner has found that 7-Eleven breached customers' privacy by collecting their sensitive biometric information without adequate notice or consent.

From June 2020 to August 2021, 7-Eleven conducted surveys that required customers to fill out information on tablets with built-in cameras. These tablets, which were installed in 700 stores, captured customers' facial images at two points during the survey-taking process -- when the individual first engaged with the tablet, and after they completed the survey.

After becoming aware of this activity in July last year, the Office of the Australian Information Commissioner (OAIC) commenced an investigation into 7-Eleven's survey.

The investigation [PDF] says 7-Eleven handled pretty much everything about this badly. It also shows the company tried to distance itself from its own tablet-based survey by blaming the third-party vendor handling the survey on its behalf.

The facial images were collected twice during the survey and stored locally on the tablets for about 20 seconds. After that, they went to the third party's servers, where they were processed and converted into an algorithmic representation of the face. The original images were then deleted from the device used to perform the survey.

These "representations" were then used to check for matches on other surveys. This was done to detect any potential gaming of the system by individuals repeatedly performing surveys and to make guesses about the age and gender of survey takers. All of that data was deleted after seven days. In total, 1.6 million surveys were performed.

7-Eleven argued this was not a violation of Australian law because the images were not used to identify, track, or monitor respondents. It also said it had no access to facial images on the local device, nor any access to images once they had been moved to the third party servers.

Wrong, says the information commissioner. The problem isn't how the collected information was handled. The problem is how it was collected. 7-Eleven needed consent from survey takers and didn't get it. The commissioner found "no evidence" individuals "expressly" agreed to have their biometric information collected by 7-Eleven.

7-Eleven argued it did get at least implied consent. As evidence of this it offered the blanket notice displayed in front of all stores:

Site is under constant video surveillance.

By entering the store you consent to facial recognition cameras capturing and storing your image.

It also pointed to its privacy policy on its website -- something survey takers weren't presented with when taking surveys.

7-Eleven may also collect photographic or biometric information from users of our 7-Eleven App and visitors to our stores, again, where you have provided your consent. 7-Eleven collects and holds such information for the purposes of identity verification.

None of this is sufficient, says the commissioner.

Consent may not be implied if an individual’s consent is ambiguous or there is reasonable doubt about the individual’s intention. While I accept that use of the tablet was voluntary, I am not satisfied that the act of using the tablet unambiguously indicated an individual’s agreement to collect their facial image and faceprint, in circumstances where:

There was no information provided on or in the vicinity of the tablet, or during the process of completing the survey, about the respondent’s collection of facial images and faceprints.

The Store Notices were unclear, and, given the prevalence of these kind of notices in stores and public places, may have created an impression that the respondent captured customers’ images using a facial recognition CCTV camera as part of surveillance of the store.

The respondent’s Privacy Policy did not link the collection of photographic or biometric information to the use of in-store ‘feedback kiosks’.

Non-specific blanket statements about possible collections are not the same thing as informing survey takers prior to taking a survey that their biometric information will definitely be collected if they fill out a survey.

That's some lawbreaking right there. The company that processed the facial images on behalf of 7-Eleven is ordered to destroy all faceprints collected by this survey. It's also forbidden from engaging in this sort of thing again without securing explicit permission from clients' customers. How much of a deterrent this is remains to be seen since the third party already declared all facial recognition data was deleted seven days after it was collected and processed.

The greater benefit of a ruling like this -- especially one that deals with information gathered irresponsibly but apparently handled with more care once it was harvested -- is the official reminder it sends to all Australian entities that may currently believe a link to a privacy policy buried on the bottom of a corporation's website home page is all that's needed to obtain "consent" for collection of personal info.

Filed Under: australia, convenience stores, facial recognition, privacy
Companies: 7-eleven

Delta Proudly Announces Its Participation In The DHS's Expanded Biometric Collection Program

from the WE-LOVE-YOU-GOVERNMENT-[girlish-screams] dept

Via Travel & Leisure comes this warning -- one the online magazine has decided to portray as exciting news.

Delta Air Lines is expanding its partnership with the Transportation Security Administration with its use of facial recognition technology making getting through airport security even quicker.

The airline is implementing a "digital identity experience" at its hub in Atlanta, offering customers with TSA PreCheck and a Delta SkyMiles number the chance to pass through security and board their flight without having to pull out a boarding pass or their ID.

Ah, the "digital identity experience." That's apparently DeltaSpeak for "biometric collection and facial recognition deployment." That's been the DHS's plan all along. It may have pretended it only wanted to post up at ports of entry (i.e., international airports) but it is going to roll this out to as many airports as possible.

How do we know "digital identity experience" is Delta PR? Because this "article" is largely a regurgitation of Delta's own press release about its increased biometric collection/domestic surveillance efforts.

First unveiled in Detroit security checkpoints in early 2021, Delta’s digital identity experience is an industry first in exclusive partnership with TSA PreCheck. The experience is expanding to Atlanta, offering customers a more efficient way to navigate the airport – without showing a paper boarding pass or a physical government ID. With just one look at a camera, customers who qualify and opt in can easily and efficiently check a bag, pass through the TSA PreCheck security line and board their plane.

You'll notice this "experience" is first being offered to PreCheck participants. This means people who have already registered and paid the government to buy back some of their Constitutional rights are the first to be invited to help expand Custom and Border Protection's biometric database.

Delta is just the "partner." The TSA just mans the front end. The data and subsequent scans belong to the CBP.

Once a customer reaches a camera at the airport, their image is encrypted and sent to U.S. Customs and Border Protection’s (CBP) facial biometric matching service via a secure channel with no accompanying biographic data. CBP then verifies a customer’s identity against government holdings and sends back an indicator to allow the customer to proceed.

There are several things wrong with all of this, starting with Travel & Leisure's cheery stenography of cheery Delta's up-sale of government intrusion. While no one expects a leisure mag to get serious about the implications of expanding biometric collections and facial recognition programs to encompass US citizens traveling domestically, the article could have used less of the Delta's GO TEAM USA jargon and more actual facts: like how facial recognition AI -- all of it -- is fundamentally flawed. Or maybe just point out the unjustified but steady encroachment of surveillance tech into airports serving millions of travelers the government has no reason to suspect are up to no good.

And Travel & Leisure may want to vet its sources. After all, Delta is more than happy to pretend things aren't the way they are as it pitches in to make traveling a worse experience for its customers.

Atlanta’s domestic terminal south security checkpoint is the first in the U.S. that will be converted to computed tomography-automated screening lane (CT-ASL) systems - making the world’s busiest airport even more efficient as travelers connect to destinations around the world.

That's simply not true. Increased efficiency may be the end goal but a recently released report by the DHS Inspector General says CT scanners are actually slowing down the screening process.

TSA deployed CT systems to airport passenger screening checkpoints that did not meet minimum throughput requirements. TSA’s February 2018 Operational Requirements Document identified the need for a CT system capable of screening, on average, 200 items per hour to successfully perform the mission. However, we determined TSA purchased 300 CT systems capable of screening an average of 170 items per hour — 15 percent less than the minimum requirement, and less than the AT X-ray system capability of approximately 354 items per hour.

This is a human centipede. The TSA says some shit. Delta Airlines reposts it with its own happy spin. And, at the tail end of it, consumer-facing sites are swallowing everything and dumping it on to web pages without bothering to question the sources or provide any information that might counterbalance the government's assertions about how more intrusive collections make America a better place to call home.

Filed Under: biometric collection, facial recognition, security theater, surveillance, tsa
Companies: delta

The Surveillance And Privacy Concerns Of The Infrastructure Bill's Impaired Driving Sensors

from the good-intentions... dept

There is no doubt that many folks trying to come up with ways to reduce impaired driving and making the roads safer have the best of intentions. And yet, hidden within those intentions can linger some pretty dangerous consequences. For reasons that are not entirely clear to me, the giant infrastructure bill (that will apparently be negotiated forever) includes a mandate that automakers would eventually need to build in technology that monitors whether or not drivers are impaired. It's buried deep in the bill (see page 1066), but the key bit is:

to ensure the prevention of alcohol-impaired driving fatalities, advanced drunk and impaired driving prevention technology must be standard equipment in all new passenger motor vehicles

The details note that the new technology should "passively monitor the performance of a driver of a motor vehicle to accurately identify whether that driver may be impaired." This isn't the kind of keylock breathalyzer test things that some people have been required to install to operate a car. This is "passive" monitoring. That means sensors and cameras. And, as Julian Sanchez highlights, that raises a ton of surveillance and privacy concerns:

This amounts to mandating a sophisticated set of sensors be installed in a space where many Americans spend huge amounts of time. (And not just commuting—many people live in vehicles, whether out of choice or necessity, at least part of the time.) A narrowly-tailored sensor that only detects blood alcohol content, if designed to immediately discard any readings below the legal threshold, might not sound worryingly invasive. But the mandate extends to monitoring for other forms of "impairment," which can require more intrusive types of sensors. One such system being developed by Nissan includes a "camera atop the instrument cluster" which "looks for facial cues signaling the driver is inebriated" while "the vehicle itself looks for driving patterns suggesting an impaired driver." In other words, one form this mandatory technology is likely to take involves pre-installed video surveillance with facial recognition capabilities. (Law enforcement, no doubt, will eagerly think of many other applications for a ubiquitous system of cameras installed in private spaces—cameras which, by design, the vehicle owner will necessarily be unable to deactivate.)

It's possible that there are versions of anti–impaired driving technology that could address these practical and privacy concerns. But that only underscores how little sense it makes for Congress to delegate authority for a regulatory mandate at a point when the technology being mandated remains largely hypothetical. Even in the absence of a mandate, there's likely to be some market for these technologies: Some drivers would embrace as a safety feature a system that warns them if they've imbibed more than they realize, or are starting to swerve on the road. They'd doubtless also be popular in contexts where the owner of a vehicle is entrusting it to another driver: Rental cars, corporate cars, commercial cab or trucking fleets. This is not, in other words, technology that will go undeveloped and untested unless it is made universally mandatory.

That makes it seem wildly premature to empower an executive branch official to mandate what is, essentially, surveillance technology in all automobiles when the precise form of the technology remains uncertain, and it's impossible to concretely debate the merits of specific systems. This sort of delegation lets legislators take credit for Doing Something to promote automotive safety and reduce the unacceptable annual death toll that results from drunk and impaired driving, without having to defend or be held accountable for any of the details of what Something entails. The bill doesn't, after all, say "we're requiring a camera that you can't shut off be installed in everyone's automobile"—that's just one possible way to "passively monitor the performance of a driver." They can reap the accolades now, and insist "that's not what we intended" if the result is a mandatory surveillance network, or cars that stop working because you've used too much hand sanitizer or a pinhole aperture in the dash got blocked.

But, let's take this even further. Sanchez notes in a parenthetical aside that law enforcement will come up with "many other applications" of this technology, but that deserves to be called out, because knowing what we know about how law enforcement embraces and "extends" every bit of surveillance technology they can get their hands on, this isn't some pie in the sky hypothetical. We all know exactly how this will play out.

Police are going to demand access to the facial recognition and other data collected by these systems, and they'll claim they need to have access to it for "public safety" reasons. Others are also likely to seek access to it as well. It's going to be an insurance goldmine.

And that's leaving aside the question of whether or not the technology will even work. The "passive" nature of it raises questions about how the system will know whether it's the driver or passengers who may be impaired. The potential to use facial recognition may raise questions about people who have facial tics or other features that the system might deem to be a sign of impairment. And, as Sanchez also notes, there's simply no clear reason why this technology needs to be government mandated and standard in every new vehicle.

Yes, preventing impaired driving is a worthy goal. But it shouldn't come at the cost of installing massive new surveillance infrastructure in every new car.

Filed Under: driving, drunk driving, facial recognition, impaired driving, infrastructure, law enforcement, privacy, sensors, surveillance

Clearview Celebrates 10 Billion Scraped Images Collected, Claims It Can Now Recognize Blurred, Masked Faces

from the getting-bigger-and-getting-worse dept

Clearview's not going to let several months of bad press derail its plans to generate even more negative press. The facial recognition tech company that relies on billions of scraped images from the web to create its product is currently being sued in multiple states, has had its claims about investigative effectiveness repeatedly debunked and, most recently, served (then rescinded) a subpoena to transparency advocacy group Open the Government demanding information on all its Clearview-related FOIA requests as well as its communications with journalists.

I don't know what Clearview is doing now. Maybe it thinks it can still win hearts and minds by not only continuing to exist but also by getting progressively worse in terms of integrity and corporate responsibility. Whatever it is that Clearview's doing to salvage its reputation looks to be, at best, counterproductive. I mean, the only way Clearview could get worse is by getting bigger, which is exactly what it's done, according to this report by Will Knight for Wired.

The company’s cofounder and CEO, Hoan Ton-That, tells WIRED that Clearview has now collected more than 10 billion images from across the web—more than three times as many as has been previously reported.

Ton-That says the larger pool of photos means users, most often law enforcement, are more likely to find a match when searching for someone. He also claims the larger data set makes the company’s tool more accurate.

That's one way of looking at it. Another way of looking at it -- and by "it," I mean Clearview's unaudited, untested facial recognition AI -- is that adding more hay increases the odds of someone grabbing some hay and thinking it's actually a needle.

Yet another way of looking at this is that Clearview's mass scraping of every bit of publicly accessible web data it can may be legal, but it certainly isn't morally acceptable. While people do largely understand that their public posts to sites can be accessed by nearly anyone, they certainly don't expect someone to collect their photos and data in bulk, package it up, and sell it to government agencies. And, in some states, this sort of activity may actually be illegal, hence the lawsuits being brought by government officials.

On top of the 10 billion images Clearview swears it will only sell to responsible adult government employees, the company is now claiming it can do some real CSI-type stuff with its tech.

Ton-That says it is developing new ways for police to find a person, including “deblur” and “mask removal” tools. The first takes a blurred image and sharpens it using machine learning to envision what a clearer picture would look like; the second tries to envision the covered part of a person’s face using machine learning models that fill in missing details of an image using a best guess based on statistical patterns found in other images.

If you feel selling government agencies a more efficient way to generate false positives and false negatives is the way to future profitability, this would be the route to take. Without a doubt, tech advances will eventually make this more accurate, but rolling out unproven machine learning on top of unproven AI is only going to compound errors. Then there's the bias problem, which has been a problem for all facial recognition software. That includes those that have been independently tested and examined by the National Institute of Standards and Technology (NIST). Notably, Clearview has yet to subject its AI to outside testing.

Finally, there's this statement from Clearview CEO Ton-That:

The company says it is not currently pitching the technology outside of the US or to private industry. “We're focusing on the United States, because we want to get it right here,” Ton-That says. “We never want this to be abused in any way.”

Whether or not this statement about its current potential customer list is true remains to be seen. Clearview has already pitched its product to private companies and foreign governments. And it appears to have exited one foreign market solely because its product was declared illegal following a government investigation.

And claiming that the company does not want its product "abused in any way" directly contradicts the stuff it says to entities it wants to sell its product to. Emails from the company's marketing staff encouraged potential law enforcement customers (as well as the occasional billionaire) to "experiment" with the software by running searches on friends, family members, and others who never consented to be part of multiple Clearview test drives.

Is Clearview the worst AI product out there? In terms of accuracy, who knows? It hasn't been independently reviewed. In terms of everything else, there's really nothing out there that competes with it. The company's nonchalant conversion of the open web into a surveillance tool sets it apart from the competition. Its latest "advances" aren't going to do anything to rehabilitate its reputation.

Filed Under: facial recognition, surveillance
Companies: clearview

Clearview Suffers Brief Bout Of Better Judgment, Drops Subpoena Demanding Activists' Communications With Journalists

from the keep-on-sucking dept

Just a few days ago, Clearview -- the company that scrapes the web to build a facial recognition database it sells to law enforcement, government agencies around the world, and a number of private parties -- decided to make itself even less likable.

It decided to subpoena transparency activists Open The Government, demanding copies of all FOIA requests it had made requesting info about Clearview. It also, more disturbingly, demanded copies of OTG's communications with journalists, clearly indicating it felt First Amendment protections were something it should enjoy, but shouldn't be extended to its critics.

It really wasn't a step Clearview needed to take… for several reasons. First of all, Clearview's reputation is pure dog shit at the moment. It's unlikely to improve unless the company pulls the plug on its product and disbands. A move like this only earns it more (deserved) disdain and hatred. What it's not going to do -- even if successful -- is deter future criticism of the business and its scraped-together facial recognition product.

Second, OTG was not a party to the lawsuit. Clearview has no right to demand these documents from a non-party, especially communications between it and journalists… journalists who are also not a party to the lawsuit Clearview is facing.

Third, if Clearview wanted copies of OTG's FOIA requests, all it had to do is visit OTG's MuckRock page and download all of its publicly accessible requests and responses.

There's some good news, though. Shortly after having shot itself in the face, Clearview had second thoughts about the self-inflicted wound it had just sustained. Here's Alexandra Levine for Politico.

Facial recognition company Clearview AI is dropping the subpoenas it served recently to some of the groups that first exposed its work with law enforcement — legal demands that sought the organizations’ correspondence with journalists and other records.

The best move would have been to never issue the subpoenas at all, especially given the First Amendment implications of attempting to pry loose communications between activists and journalists. This withdrawal is a distant second on the list of "Smart Things to Do."

And here's the rationale for the decision, straight from the mouth of Clearview's founder:

Clearview CEO Hoan Ton-That told POLITICO that “on further reflection about the scope of the subpoenas, and my strong view of freedom of the press, we have decided to withdraw the subpoenas served on Open the Government and its associates.”

I imagine that "reflection" showed a badly maimed face mouthing the word "why" after taking the full impact of an unforced error. This should have been the first response to the suggestion of subpoenaing non-parties and demanding copies of their conversations with journalists. It should have been offered by Clearview's legal team. And if those bright minds failed to comprehend the collateral damage this move would inflict on the First Amendment (not to mention the fact that Clearview would be sitting at ground zero when its own bomb went off), the genius who came up with the idea of treating millions of internet users as fodder for unproven facial recognition tech should have been able to see doing this was a terrible idea.

Then again, it may just be impossible for Clearview and its principals (note the spelling -- "principles" are different and not possessed by Clearview) to get a firm grasp on the obvious if it can't easily be scraped from content created by people with far more common sense than its founder and legal reps.

Filed Under: activism, facial recognition, foia, intimidation, journalism, subpoena, surveillance, transparency
Companies: clearview, open the government

GAO's Second Report On Facial Recognition Tech Provides More Details On Federal Use Of Clearview's Unvetted AI

from the still-greater-than-zero-agencies,-unfortunately dept

A couple of months ago, the Government Accountability Office completed the first pass of its review of federal use of facial recognition technology. It found a lot to be concerned about, including the fact that agencies were using unproven tech (like Clearview's ethical nightmare of a product) and doing very little to ensure the tech was used responsibly.

Some agencies appeared to have no internal oversight of facial recognition tech use, leading to agencies first telling the GAO no one was using the tech, only to update that answer to "more than 1,000 searches" when they had finished doing their first pass at due diligence.

A more complete report [PDF] has been released by the GAO, which includes answers to several questions asked of federal agencies using the tech. Unfortunately, it confirms that many agencies are bypassing what little internal controls are in place by asking state and local agencies to run searches for them. DHS entities (CBP, ICE) did the most freelancing using downstream (governmentally-speaking) databases and tech.

For whatever reason, CBP and ICE (which have access to their own tech) are using agencies in Ohio, Nebraska, Michigan, Kansas, and Missouri (among others) to run searches for criminal suspects and to "support operations." A whole lot of non-border states are allowing agencies to bypass internal restrictions on use of the tech.

And there's a whole lot of Clearview use. Too much, in fact, considering the number of agencies using this highly questionable product exceeds zero.

The US Air Force says it engaged in an "operational pilot" beginning in June 2020, utilizing Clearview to run searches on biometric information gathered with "mobile biometric devices, including phones."

The Inspector General for the Department of Health and Human Services also apparently used Clearview. The report says the HHS OIG "conducted an evaluation of the system in an attempt to identify unknown subjects of a criminal investigation." Experimentation, but with the added bonus of possibly infringing on an innocent person's life and liberty!

Also on the list are CBP, ICE, and US Secret Service. ICE appears to be the only agency actually purchasing Clearview licenses, spending a total of $214,000 in 2020. The CBP, however, is getting its Clearview for free, utilizing the New York State Intelligence Center's access to run searches. The Secret Service gave Clearview a test drive in 2019 but decided it wasn't worth buying.

The Department of the Interior says it has both stopped and started using Clearview. Under "Accessed commercial FRT [facial recognition technology] system, the DOI claims:

Interior uses Clearview AI to verify the identity of an individual involved in a crime and research information on a person of interest. Interior may submit photos (e.g., surveillance photos) for matching against the Clearview AI’s repository of facial images from open sources. U.S. Park Police reported it stopped using Clearview AI as of June 2020.

But under "New access to commercial FRT system," the DOI states:

Interior reported its U.S. Fish and Wildlife Service began using a trial version of Clearview AI in May 2020, and purchased an annual subscription in June 2020.

The DOI is both a current and former customer, depending on which component you speak to, apparently.

The DOJ is an apparent believer in the power of Clearview, providing access to the ATF, DEA, FBI, and US Marshals Service. But there must be a lot of sharing going on, because the DOJ only purchased $9,000-worth of licenses.

Interestingly, the DOJ also notes it received an upgrade from Axon, which provides body-worn cameras. Axon has apparently added a new feature to its product: "Facial Detection." Unlike facial recognition, the product does not search for faces to run against a biometric database. Instead, the system "reviews footage" to detect faces, which can then be marked for redaction.

This FRT-related expenditure is also interesting, suggesting the DOJ may actually be trying to quantify the effectiveness of body cameras when it comes to deterring officer misconduct.

DOJ reported that it awarded an $836,000 grant to the Police Foundation for the development of techniques to automate analysis of body worn camera audio and video data of police and community interactions. In particular, these techniques could (1) allow an evaluation of officers’ adherence to principles of procedural justice and (2) validate the ratings generated by the automated process using a randomized control trial comparing software ratings of videos to evaluations performed by human raters under conditions of high and low procedural justice.

Finally, there's this unnecessarily coy statement by the IRS about its use of commercial facial recognition systems.

A third-party vendor performed facial recognition searches on behalf of the IRS for domestic law enforcement purposes. Additional details on the search are sensitive.

Whatever. It's probably Clearview. And if it isn't, it probably will be at some point in the near future, given federal agencies' apparent comfort with deploying unproven, unvetted tech during criminal investigations.

The report is probably the most comprehensive account of facial recognition tech by the federal government we have to work with at the moment. It shows there's a lot of it being used, but it hasn't become completely pervasive. Yet. Most agencies use the tech to do nothing more than identify employees and prevent unauthorized access to sensitive areas. Some agencies are digging into the tech itself in hopes of improving it. But far too many are still using a product which has been marketed with false statements and has yet to have its accuracy tested by independent researchers. That's a huge problem, and, while it's not up to the GAO to fix it, the report should at least make legislators aware of an issue that needs to be addressed.

Filed Under: doi, doj, facial recognition, gao
Companies: clearview

US Army Now Using Clearview's Unproven Tech To Investigate Crimes

from the sole-source-provider-of-a-service-no-one-else-is-willing-to-create dept

We can add another government agency to the list of entities that have been suckered in by Clearview's highly questionable sales pitches about its unproven tech: the US Army. [Paywall ahead, but alternatives abound.]

The US Army has a contract with Clearview AI, according to documents that reveal the controversial facial-recognition startup making bold claims to the military about capabilities such as "criminal network discovery" and "force protection and area security."

The contract, obtained with other documents by Insider via public-records request, shows the US military awarding a discounted contract for Clearview to work with the Army's Criminal Investigation Command, which investigates serious crimes that could involve active service members or civilian workers for the Army.

Fantastic. Now, let's back up a bit to the opening paragraph. By "highly questionable," I mean many of Clearview's assertions about its crime fighting efficacy have been directly contradicted by law enforcement agencies who've test-driven the product. And by "unproven," I mean Clearview has never submitted its facial recognition AI to outside, independent testing -- like that performed by the NIST. And by "suckered in," I mean the Army fell for all of it, converting its test run of Clearview's tech into a sole-source contract for fifteen licenses.

The memo [PDF] obtained by Business Insider says Clearview is one of a kind.

Clearview Al provides unique and innovative capability with regard to facial recognition technology and its use by law enforcement organizations. The 502d MP BN (CID) has been using this tool on a trial basis and has seen an increase of 15%- 25% success in positively identifying potential subject, victims and witnesses in possible crimes under Army CID jusridiction. They were the only company capable of providing the service.

I'm not sure about that last sentence. It would read better as "the only company willing to provide the service." Unlike other, more reputable facial recognition tech providers, Clearview's database of facial photos and personal information is scraped from thousands of public websites and social media services. It isn't the only company capable of doing this. But so far, it's been the only one willing to scrape the web and sell access to its billions of harvested images.

That may make Clearview unique. But it doesn't make it good or reputable or tolerable. And if the CID is seeing more positive identifications, it's only because it has more images to work with. What it doesn't have is a (excuse the term) battle-tested tech solution.

The claims made by the Army and the claims made by Clearview to the Army haven't been supported by any evidence from either party. Multiple requests for comments, clarifications, or supporting evidence went ignored by both the Army CID and its preferred facial recognition tech provider. Clearview also declined to explain what its marketing material [PDF] means when it claims its AI can help with "force protection and area security" or assist investigators in uncovering "criminal networks."

It also didn't explain this claim, which was almost immediately refuted by the party namechecked in the bullet point.

Later in the flyer, Clearview says that it's been "rated 99.6% accurate" per an accuracy benchmark created by the University of Washington's MegaFace image dataset. This claim hasn't been independently verified by any third party.

"We don't know how they tested their software, and we haven't evaluated their algorithms," Ira Kemelmacher-Shlizerman, former head of the MegaFace challenge, told Insider. "The MegaFace challenge has been closed for a while now, and no one on our teams is working with it."

This tracks with Clearview's previous accuracy assertions. It makes claims but refuses to allow any outside testing of its algorithm. Until it's willing to do that, there's no reason anyone -- much less the investigative wing of the US Army -- should test drive its tech, much less purchase fifteen licenses. The CID is playing with fire here, and those overseeing this division should be demanding answers from Clearview, and blocking access to its tech until it has something more solid than the company's questionable accuracy claims to work with.

Filed Under: army, facial recognition, us army
Companies: clearview

Detroit Skating Rink Horns In On Detroit PD's Facial Recognition Gaffe Racket, Denies Teen Girl Opportunity To Skate

from the even-if-our-software-is-wrong,-we-will-still-enforce-its-decision dept

It looks like Detroit, Michigan is trying to corner the market on bad facial recognition tech. The city's police department is already associated with two false arrests based on bad matches by facial recognition software. This latest news, via Techdirt reader Jeffrey Nonken, shows mismatches aren't just limited to the public sector.

A Black teenager in the US was barred from entering a roller rink after a facial-recognition system wrongly identified her as a person who had been previously banned for starting a fight there.

Lamya Robinson, 14, had been dropped off by her parents at Riverside Arena, an indoor rollerskating space in Livonia, Michigan, at the weekend to spend time with her pals. Facial-recognition cameras installed inside the premises matched her face to a photo of somebody else apparently barred following a skirmish with other skaters.

The teen told staff it couldn't possibly be her since she had never visited the venue before. But it didn't matter to the management of the rink, which is located in a Detroit suburb. She was asked to leave and now her parents are considering suing the rink over the false positive. Fortunately, no one at the rink felt compelled to call the police, which likely wouldn't have helped anything considering local law enforcement's track record with faulty facial recognition search results.

As for Riverside Arena, it's apologetic but not exactly helpful. Management claims deploying facial recognition tech on patrons is part of the "usual" entry process. It also unexplained that it's "hard to look into things when the system is running," which I suppose means that's why no one could double-check the match while Robinson was still there contesting the search results.

Being wrong some of the time is also good enough for non-government work, apparently.

"The software had her daughter at a 97 percent match. This is what we looked at, not the thumbnail photos Ms. Robinson took a picture of, if there was a mistake, we apologize for that."

Obviously, there was a mistake. So, this should have just been an apology, not a half-hearted offer of an apology if, at some point in the future, someone other than the people directly affected by this automated decision steps forward to declare the mismatch a mismatch.

This is the other side of the facial recognition coin: private sector use. This is bound to result in just as many mismatches as government use does, only with software that's perhaps undergone even less vetting and without any direct oversight outside of company management. False positives will continue to be a problem. How expensive a problem remains to be seen, but since private companies are free to choose who gets to use their services, lawsuits probably won't be much of a deterrent to deploying and using unvetted software that will keep the wrong people out and, perhaps more disturbingly, let the wrong people in.

Filed Under: detroit, facial recognition, livonia, michigan, skating rink
Companies: riverside arena

Texas School District's Facial Recognition System Capable Of Capturing A Single Student's Image More Than 1,000 Times A Week

from the Panopticon-ISD dept

Facial recognition tech is making its way into schools, subjecting minors to the same tech that still hasn't proven its worth in the adult world. Like many other surveillance encroachments, this acquisition and deployment was prompted by violence and fear.

Alfred Ng of The Markup has obtained documents detailing a system in use in some Texas schools, one acquired as a potential answer to a uniquely American existential threat.

The school district originally purchased AnyVision after a mass shooting in 2018, with hopes that the technology would prevent another tragedy. By January 2020, the school district had uploaded 2,967 photos of students for AnyVision’s database.

This seemingly small number only covers a few months of use by the Santa Fe (Texas) Independent School District. Prior to its official rollout, the district ran a short test using an even larger set of photos.

With more than 5,000 student photos uploaded for the test run, AnyVision called the results “impressive” and expressed excitement at the results to school administrators.

“Overall, we had over 164,000 detections the last 7 days running the pilot. We were able to detect students on multiple cameras and even detected one student 1100 times!” Taylor May, then a regional sales manager for AnyVision, said in an email to the school’s administrators.

It seems like the perfect tool to track students individually as they go about their school day. The 1,100 hits for a single student is considered a sign of the software's effectiveness, something administrators seem unhealthily enthusiastic about.

And that's not all the district did with its new tech. It also lent it to local law enforcement officers who were trying to identify a suspected drug dealer they believed was a student. The school's cops contacted AnyVision which uploaded the provided photos and ran it against images captured by its system.

The company says the software is watchlist-based, limiting users to targeting persons of interest. Private retailers use it to keep shoplifters and other criminals out of stores. Schools can target suspected sex offenders by using publicly available mugshots from law enforcement databases.

But this trial run suggests the district isn't interested in limiting itself to watchlists. The user guide [PDF] obtained by The Markup makes it clear the software can log all faces that pass by equipped cameras. It's this kind of broad collection that enables more than 100,000 "detections" in a single week and allows students to be pinpointed 1,000 times during a five-day school week.

There are controls available in the software, but it's not clear the district is using any of them.

The software offers a “Privacy Mode” feature in which it ignores all faces not on a watchlist, while another feature called “GDPR Mode” blurs non-watchlist faces on video playback and downloads. The Santa Fe Independent School District didn’t respond to a request for comment, including on whether it enabled the Privacy Mode feature.

As for AnyVision, it says hey, it's just the provider.

“We do not activate these modes by default but we do educate our customers about them,” AnyVision’s chief marketing officer, Dean Nicolls, said in an email.

This lack of clarity or comment suggests the district is using the tech for all-purpose surveillance rather than utilizing it in the targeted fashion idealized in public statements and marketing brochures. The public statements by the district's board claim there's no invasion of privacy here, just another layer of protection. While the first part of the statement may be technically true given the lack of an expectation of privacy on school grounds, the latter is much more difficult to quantify as a justifiable offset for always-on surveillance.

Filed Under: anyvision, facial recognition, schools, surveillance, texas