On September 22, 2014, an anonymous activist nearly brought down the Seattle Police Department. By typing a few lines on his keyboard, the hacker threatened to disrupt policing, derail a technology upgrade, annoy the federal government, and send costs spiraling. “It was like a DDOS attack,” remembers Mike Wagers, Seattle PD’s Chief Operating Officer. “It freaked everybody out that this was going to shut the system down.”
But this was no traditional hacking attack: no firewalls were breached, no files were stolen. The activist, known only by his email handle, email@example.com, had made a simple demand. He wanted every video ever captured by Seattle police car dashboard cameras, as well as all the videos from a new body camera pilot project. “I’m requesting these records for doing my own analysis for picking videos to request,” the man wrote in an email obtained by The Seattle Times. Later, he told me, “I just wanted to watch and let other people watch footage.”
In most parts of the world, such a demand would be met by a polite refusal. But Washington state’s public disclosure laws allow requests that are both anonymous and massively broad. The activist’s request encompassed over 360 terabytes of data: 1.6 million recordings stretching back six years. Each one would have to first be reviewed to cut out any images or audio of children, medical or mental health incidents, confidential informants, or victims or bystanders who did not want to be recorded.
Using the Department’s existing set-up, where footage stored on DVDs was manually edited on a laptop, redaction took about an hour for every minute of video on file. Considering that most clips are at least 10 minutes long, firstname.lastname@example.org’s demand represented thousands of person-years, and hundreds of millions of dollars, of work. “There was no way we could redact that much data within our lifetimes,” says Wagers.
Faced with a similar request from the mysterious activist, another Washington police department, in Bremerton, swiftly shelved its plans to put body cams on the streets. But the political fall-out from such a move in Seattle would be devastating. In 2012 the Department of Justice had found that the Seattle Police Department (SPD) had engaged in a pattern of excessive force and discriminatory policing that violated the Constitution and federal law. It placed the SPD under a five-year plan of reform that included far more dashboard and body cameras. In December the DOJ warned the SPD that in-car video and body cameras are “necessary and vital evidentiary records that must be available to resolve controversies and increase community trust.”
Seattle is hardly unique in its policing controversies or its growing troves of sensitive data. After civil unrest following the recent death of a black man in police custody, Baltimore Mayor Stephanie Rawlings-Blake called on the Department of Justice to investigate the city’s police department for “unconstitutional practices” including excessive force and discrimination. Rawlings-Blake also promised to implement police body cameras by the end of the year. Chicago, Los Angeles, New York and Philadelphia are also testing the devices — and struggling with how to handle the footage.
Caught between the devil of federal anger and the deep blue sea of financial ruin, the SPD seemed to be out of options. And so Wagers did what any public official does when faced with an impossible technical challenge and no budget: he threw a hackathon. “We had this enormous request that we couldn’t possibly fulfill with current technology,” he tells me. “What did we have to lose?”
Police body cameras are a relatively new phenomenon. Steve Ward claims to have come up with the idea as a Seattle mountain bike cop in 1997. A photographer working on a story threw a camera around Ward’s neck and set it to shoot a stream of pictures as Ward hurtled down a hill near the city’s touristy Pioneer Square. “Dashcams were just starting to catch on and I thought, what are we doing? We need to put the camera on the officer, not on the dashboard,” he says.
Ward is now CEO of VieVu, a Seattle company that has sold body cams to over 4000 law enforcement agencies in 16 countries. “Without a doubt, every cop in America is going to be wearing a body camera in the next couple of years,” he says. “If I was still working the street, I would want a body cam. If anybody makes any accusation against me — let’s go to the video.”
Going to the video can work. In South Carolina in April, Walter Scott, an unarmed black man, was shot dead by a white police officer who claimed that Scott had grabbed his Taser. Cellphone footage from a passerby, however, showed the officer shooting Scott in the back and then apparently planting the Taser on his body. The officer was charged with murder, and a bill requiring police body cameras is now working its way through South Carolina’s legislature. Similar bills are on the table in Iowa, Ohio and Nevada, and Presidential candidate Hilary Clinton recently called for body cameras to be the “norm everywhere.”
The little research that has been done into the effect of body cameras on police misconduct suggests they can have a positive effect. A year-long study in Rialto, California, found that police officers who wore body cameras had half as many ‘use of force’ incidents during their shifts as those that did not. The researchers also tried to track changes in complaints against officers over the year but couldn’t — the reduction was so large that there weren’t enough complaints to analyze.
An earlier, more limited request by email@example.com for body cam videos had convinced him of their value in overseeing rogue cops. “One agency hand-picked [a video of] them Tasering a suspect,” he says. “I had concerns about that. By the time the person was Tased, he had surrendered. So I did uncover something, and I was surprised because it was a hand-picked video.”
Ward argues that collecting video now has wide support. “But everybody differs a little bit on what to do with it after that.” The same cameras that improve transparency and accountability can also violate the privacy of people in front of and behind the lens. Police officers don’t want to be recorded bitching about their superiors. Witnesses and informants are understandably wary about who gets to see the recordings, and protesters worry about being tracked by government agencies.
Policies about where and when to turn cameras on, language to warn people who are being filmed, and limits on using the footage in investigations can address some of these concerns. But liberal public disclosure laws like Washington’s leave a gaping loophole. How can police departments release videos to an eager public without invading the privacy of victims, patients and bystanders on some of the worst days of their lives?
On November 20, Mike Wagers announced the SPD’s hackathon on Twitter and promised to make sample dashcam videos available. A week before the hackathon, a local newspaper had obtained (ironically, through an SPD public records request) text messages sent from the activist’s cellphone, and got in touch. “They might have been able to determine my name from the phone number,” he says. “The newspaper convinced me to release my name.”
And so firstname.lastname@example.org unmasked himself as Timothy Clemans, a 24-year-old programmer whose previous claim to fame was holding the record for the most consecutive daily visits to Seattle’s Space Needle (at least 113). His first public act was to tweet Mike Wagers an offer of assistance with the hackathon — as long as he could talk with whoever was currently doing redaction.
It would be difficult to come up with a more stereotypical hacker than Clemans. He was home-schooled. He painstakingly taught himself HTML at the age of 8. And he does indeed live in his parent’s basement in the suburbs. Clemans has struggled with depression in the past, and credits his Space Needle project with changing him from “a depressed, shy and lazy kid to a happy, out-going, disciplined young man.” Then last summer, his focus changed. A local TV station had gone to court to access Seattle’s police dashcam videos. “I was very irritated that the TV station only aired clips and only cared about Seattle,” he remembers. “I just wanted all the footage.” Clemans had a new mission.
“We met Timothy the very next day, with the people around the table that he had requested,” says Wagers. “The kid’s extremely bright. We said drop your request, we’ll show the limitations we have in terms of video redaction and maybe you could help us.”
As members of the department described their painstaking methods, Clemans couldn’t believe what he was hearing. “It really did surprise me that the police department was not prepared. For agencies to not have a plan for dealing with this stuff was pretty shocking. I thought they would have an efficient way of getting videos and reviewing them. It’s a failure of imagination.”
Clemans had been thinking about the idea of ‘over-redaction’: applying digital filters to videos en masse to obscure people’s faces, locations and embarrassing items, such as sex toys in people’s homes. In the week before the hackathon, he hurriedly put together a script that stripped out the audio, blurred each video frame, turned colors into grayscales, and shrank the file size by about 75 percent.
Wagers knew he would have at least one entry to the hackathon, but how many others in left-leaning Seattle would help a police department with a reputation for violence and discrimination? He needn’t have worried. On the day of the event, December 19, over 80 people crowded into a basement at police headquarters to show off their solutions. “It was straight out of central casting,” says Wagers. “There were people in the room who had sued us before over public disclosure — and won. There were people in there that I had seen on the protest lines for Black Lives Matter. And there were others who you would think would come to hackathon.” (Yes, he means geeks).
An independent programmer and a team of students from the University of Washington demonstrated facial blurring using open source tools, while local companies, including Microsoft and Evidence.com (a division of Taser that stores video for thousands of police departments), touted their commercial offerings. But it was ultimately Clemans’s over-redaction script that had the most support. “Tim’s idea carried the day,” says Wagers. “Everyone said he had the best code.”
Clemans signed an unpaid research agreement with the SPD and, assisted by other volunteer programmers, worked on refining his over-redaction script. The trick would be to blur images enough so that individuals could not be recognized, and yet retain enough detail to alert viewers if something fishy was going on — say, an unreported incident of use of force by an officer. Citizens could then request just that part of the video, which would be manually redacted and disclosed as before.
In January an opportunity to start sharing videos presented itself. During a march to celebrate Martin Luther King Jr Day, police and protesters scuffled. “People were interested in seeing that, so we ran the code on six or eight hours of video from body cams and posted it on YouTube,” Wagers says. “We realized at the time that the code was not quite there yet but we were trying to be as transparent as possible.”
One of the people who watched the MLK Day videos was Jared Friend, Technology and Liberty Director at the American Civil Liberties Union of Washington. He was unimpressed. “I was able to identify four people that I knew,” he says. “Video has an enormous amount of contextual information. You may still be able to make out buildings, and if you know the video is of a particular person, all of a sudden you know where they may be at a certain time of day.”
Clemans was also unhappy with the blurring. “We’re trying to figure out which over-redaction approach we’re going with,” he says. “We’ve actually gone away from blurring to just showing the outlines of lines.” In late April, he released the first videos redacted with the new filter, which makes it easier to see what is happening but keeps faces hidden.
Clemans is also working on a smarter solution for audio than just muting it. He is now using Microsoft Media Indexer, an online voice recognition tool that can produce transcriptions of spoken language. By checking every word of a transcription against a dictionary, his software can screen for names and remove audio only where it recognizes a name. “It won’t be totally automated anytime soon because the recognition is just not reliable enough, but it’s still useful,” Clemans says.
Running the current version of Clemans’s over-redaction software on the cheapest servers in Amazon’s cloud takes around half the duration of the video being processed (compared to about 60 times the video’s length when done manually). Ultimately, Clemans would like a more sophisticated script that blurs only what needs to be blurred, automatically tracking and obscuring faces and license plates. “If we get this detection stuff figured out, where we can redact all faces, even the officers, I think the longest it might take is twice the length of the video.”
Not everyone at the SPD has welcomed the wunderkind activist, however. Some members of the department kept their distance from the person who only months before had posed a threat. “We were having troubles with the research agreement,” says Clemans. “I wasn’t really being given the materials that I needed.”
Wagers was also becoming worried about Clemans himself. “I found out a couple of months ago he was on welfare,” he says. “So I thought about it. He’s been helping us since last year and has saved us probably millions of dollars in terms of public disclosure. We came up with a plan to hire him on a temporary, three-month contract.”
The $22-an-hour position should raise Clemans’s standing within the department, and give him the freedom to do what he wanted in the first place — watch and let other people watch the footage. “They felt hiring me would make it a bit more clear to people that they should work with me,” says Clemans. “But I’m not totally sure how this is all going to turn out.” When I asked Wagers whether employing Clemans was a tough sell within the SPD, he went silent. “Let’s just say it was a non-conventional hire. The chief supported it,” he says eventually. “Let’s just leave it at that.”
Clemans and Wagers expect the SPD’s over-redaction system to be complete sometime this year. Their vision is of an officer simply docking her body cam at the end of a shift. The footage would then be automatically uploaded to storage, either locally or in the cloud, over-redacted for privacy and posted online for everyone to see within a day. Sensitive recordings would be withheld and flagged for full redaction, ideally by picking out keywords such as ‘rape’ or ‘juvenile’ from the audio stream or the computerized dispatch system. “We don’t want the officer to have to think about the technology at all,” says Wagers.
Such an automated assembly line worries others. “This whole thing is an interesting experiment but it’s futile because someone like Tim Clemans could still come along and request all the unredacted videos,” says the ACLU’s Jared Friend. “The Seattle PD hopes that by proactively posting these videos it will reduce the likelihood of that, but I’m a bit skeptical.”
The bigger issue for Friend is ensuring that public disclosure laws exclude thrill-seekers and concentrate on reducing police brutality. “Redaction isn’t the solution,” he says. “Much better would be limiting disclosure of videos for the purpose for which these cameras exist: cases where someone can point to facts or suspicions of police misconduct.” He doesn’t think any police department has struck the right balance yet.
Nationally, a few jurisdictions would like to exclude body camera footage from public records laws altogether — although such moves are almost certain to be challenged in court. Even public interest groups can’t agree. Last week, a coalition of over 30 civil rights, privacy and media organisations including the ACLU, the NAACP and the Electronic Frontier Foundation issued five principles on body worn cameras that included only the vague phrase: “Departments must consider individual privacy concerns before making footage available to broad audiences.”
For the foreseeable future, agencies such as the Seattle Police Department will have to continue to walk a fine line between the public’s thirst for access and the public’s fears for their privacy.
Wager’s frustration with the system is obvious: “We have to figure out a way to release it while at the same time protecting privacy rights — but we can’t pick and choose the video we’re publishing, which is essentially what we’re doing now.” So far, the SPD has succeeded in turning one troublesome activist into an asset. The next one might prefer to beat them rather than join them.