Homeland Security Worries Covid-19 Masks Are Breaking Facial Recognition, Leaked Document Shows

While doctors and politicians still struggle to convince Americans to take the barest of precautions against Covid-19 by wearing a mask, the Department of Homeland Security has an opposite concern, according to an “intelligence note” found among the BlueLeaks trove of law enforcement documents: Masks are breaking police facial recognition.

The rapid global spread and persistent threat of the coronavirus has presented an obvious roadblock to facial recognition’s similar global expansion. Suddenly everyone is covering their faces. Even in ideal conditions, facial recognition technologies often struggle with accuracy and have a particularly dismal track record when it comes to identifying faces that aren’t white or male. Some municipalities, startled by the civil liberties implications of inaccurate and opaque software in the hands of unaccountable and overly aggressive police, have begun banning facial recognition software outright. But the global pandemic may have inadvertently provided a privacy fix of its own — or for police, a brand new crisis.

A Homeland Security intelligence note dated May 22 expresses this law enforcement anxiety, as public health wisdom clashes with the prerogatives of local and federal police who increasingly rely on artificial intelligence tools. The bulletin, drafted by the DHS Intelligence Enterprise Counterterrorism Mission Center in conjunction with a variety of other agencies, including Customs and Border Protection and Immigration and Customs Enforcement, “examines the potential impacts that widespread use of protective masks could have on security operations that incorporate face recognition systems — such as video cameras, image processing hardware and software, and image recognition algorithms — to monitor public spaces during the ongoing Covid-19 public health emergency and in the months after the pandemic subsides.”

The Minnesota Fusion Center, a post-9/11 intelligence agency that is part of a controversial national network, distributed the notice on May 26, as protests were forming over the killing of George Floyd. In the weeks that followed, the center actively monitored the protests and pushed the narrative that law enforcement was under attack. Email logs included in the BlueLeaks archive show that the note was also sent to city and state government officials and private security officers in Colorado and, inexplicably, to a hospital and a community college.

The new public health status quo represents a clear threat to algorithmic policing.

Curiously, the bulletin fixates on a strange scenario: “violent adversaries” of U.S. law enforcement evading facial recognition by cynically exploiting the current public health guidelines about mask usage. “We assess violent extremists and other criminals who have historically maintained an interest in avoiding face recognition,” the bulletin reads, “are likely to opportunistically seize upon public safety measures recommending the wearing of face masks to hinder the effectiveness of face recognition systems in public spaces by security partners.” The notice concedes that “while we have no specific information that violent extremists or other criminals in the United States are using protective face coverings to conduct attacks, some of these entities have previously expressed interest in avoiding face recognition and promulgated simple instructions to conceal one’s identity, both prior to and during the current Covid-19 pandemic.” This claim is supported by a single reference to a member of an unnamed “white supremacist extremist online forum” who suggested attacks on critical infrastructure sites “while wearing a breathing mask to hide a perpetrators [sic] identity.” The only other evidence given is internet chatter from before the pandemic.

But the bulletin also reflects a broader surveillance angst: “Face Recognition Systems Likely to be Less Effective as Widespread Wear of Face Coverings for Public Safety Purposes Continue,” reads another header. Even if Homeland Security seems focused on hypothetical instances of violent terrorists using cloth masks to dodge smart cameras, the new public health status quo represents a clear threat to algorithmic policing: “We assess face recognition systems used to support security operations in public spaces will be less effective while widespread public use of facemasks, including partial and full face covering, is practiced by the public to limit the spread of Covid-19.” Even after mandatory mask orders are lifted, the bulletin frets, the newly epidemiologically aware American public is likely to keep wearing them, which would “continue to impact the effectiveness of face recognition systems.”

The battle over masks predates the pandemic. During the 2011 Occupy Wall Street protests, the New York City Police Department pulled legal gymnastics to arrest demonstrators donning grinning Guy Fawkes masks popularized by the hactivist group Anonymous, citing an 1845 law that bans groups of two or more people from covering their faces in public except at “a masquerade party or like entertainment.” Bans in states around the country followed, often in response to protest movements. In 2017, for example, North Dakota banned masks amid protests over the Dakota Access pipeline. Other anti-mask laws were designed to prevent Ku Klux Klan gatherings, though often with the primary aim of protecting white elites.

The Homeland Security document cites as cause for concern tactics used in recent pro-democracy demonstrations in Hong Kong. In that movement, which coincided with the emergence of China’s sophisticated surveillance state, police carried around cameras attached to poles, presumably to capture the faces of protesters. Demonstrators responded by shining laser pointers at police, sawing down lampposts mounted with cameras, and masking up. “At first only the militants wore masks,” Chit Wai John Mok, a Ph.D. student in sociology who studies social movements, told The Intercept. “But later on when even peaceful assemblies or marches were also banned, most protesters, moderates or militants, wore them.”

In Hong Kong, too, authorities saw masks as a problem. Last October, Chief Executive Carrie Lam banned face coverings at protests, further enraging protesters. In January, as Covid-19 spread, the government abruptly reversed its policy and began encouraging people to wear masks in public places.

In the past few months, companies around the world have scrambled to adapt their systems to facial coverings, with a few claiming that they can identify masked faces. So far, there is little evidence to support these claims. Some companies appear to have updated their algorithms by photoshopping masks onto images from existing datasets, which could lead to significant errors. To use facial recognition to identify individuals on the street, “it would be best to have lots of real life examples showing the many ways people wear masks and the different angles they get captured,” Charles Rollet, an analyst with IPVM, an independent group that tracks surveillance technology, told The Intercept. Without such images, he added, “There’s a risk of a substantially higher false positive rate, which, in a law enforcement setting, could lead to wrongful arrests or worse.” IPVM tested four facial recognition systems in February and found that their performance was drastically reduced with masked faces.

Homeland Security’s Customs and Border Protection, which uses facial recognition screening on international travelers, has also claimed that its technology works on masked faces. In that scenario, however, travelers look straight into the camera — an angle that makes it easier to identify them, even with masks.

Homeland Security has recently come under fire for efforts to expand the use of facial recognition by CBP. In December, following public outcry, department officials walked back plans to make facial recognition of U.S. citizens mandatory in airports when they fly to or from international destinations. Current protocols allow citizens to opt out of facial recognition screening.

Even as Homeland Security warned in the document of the ostensible risks posed by masked “violent adversaries,” the agency cautioned about violence perpetrated by anti-maskers. The same day that the Minnesota Fusion Center distributed the intelligence note, it circulated a second one warning that some people viewed mask orders as “government overreach” — and would sooner fight than cover their faces. “There have been multiple incidents across the United States,” the second document read, “of individuals engaging in assaults on law enforcement, a park ranger, and essential business employees in response to requests to wear face masks and to abide by social distancing policies.”