by Michael Dean Thompson

Social media has revolutionized connectivity, allowing people to develop and maintain relationships well beyond what was possible just a generation earlier. The revolution has, however, enabled the joint planning, execution, and documentation of crimes. It is therefore no surprise that law enforcement wants to access social media to help them investigate those crimes.

Cops are always looking for means to bypass the need to obtain warrants. Given the chance, they would prefer to perform bulk surveillance rather than submit to the limits and transparency of a warrant. Ironically, even as cops strive to uncover the activities of people not yet suspected of any crimes, they wish to obscure their own practices from both the public and the courts.

To that end, police agencies have turned to software solutions capable of scraping vast quantities of data from social media platforms. Algorithms within the software can then create reports detailing groups, people, and topics that interest police intelligence officers. The algorithms can even go further by drawing inferences regarding target viewpoints and associations. These tools work far faster and at a lower cost than police officers while enabling a near omniscient viewpoint into the lives of the surveilled accounts. The Brennan Center for Justice’s report, “Principles for Social Media Use by Law Enforcement,” warns that this stands in contrast to the U.S. Supreme Court’s recognition that the drafters of the Constitution wanted “to place obstacles in the way of a too permeating police surveillance.” Carpenter v. United States, 138, S. Ct., 2206 (2018).

Biased Policing and Surveillance

Social media seems poised to exacerbate racially biased policing. Rachel Levinson-Waldman points out in a 2019 Howard Law Journal article that “youth of color are particular targets, with the most high-profile examples arising in the context of gang surveillance, raising concerns that already over-policed communities will bear the brunt of its intrusion.” 61 How. L. J. 523. Yet, because the source of the intelligence may never be transcribed into the investigation’s record, the true depth of the biased policing may never come to light.

Gang databases, which are disproportionately populated by people of color, are often supplemented by social media scraping. These databases are notorious for their inaccuracies, yet they are sometimes used to increase bail, escalate the encounter during traffic stops, and lengthen sentences. The city of Chicago Officer of Inspector General noted in 2019 that 95% of their gang database’s membership were people of color even as there was a lack of transparency as to how the list was built. To highlight the challenge of such a lack of transparency, the NYPD’s inspector general noted that people could be added to their gang database solely based on social media content and “self-admission” or “an individual’s use of language, symbols, pictures, or colors associated with a criminal group.” Surprisingly, this can include just being found in a photograph with gang members.

Police may determine how to respond to calls using similar means. The Fresno Police Department in California uses a tool called Beware to draw on social media information and create a “threat score.” Dispatchers can then pass that information on to cops who are sent out on calls. Those officers would then potentially arrive more anxious or primed for violence or even accompanied by SWAT. It is not difficult to see how erroneous scores in either direction might amplify danger for both civilians and police.

All too often, police have shown a preference for tracking racial justice advocates. In general, they do this through the blunt instrument of keyword searches. As an example, the Boston Police Department used a social media monitoring tool to track mentions of Black Lives Matter and Muslim Lives Matter. The Los Angeles Police Department searched for the hashtag #BlackLivesMatter and #SayHerName as well as the names of people who had been killed by police, like Sandra Bland and Tamir Rice. In this case, just being committed to activism is enough to place a person squarely under police attention and surveillance.

In 2020 during the racial justice protests over the killings of George Floyd and Breonna Taylor, Dataminr’s social media monitoring tool helped law enforcement agencies across the country monitor protestors via their social media presence. Dataminr is a company that provides an AI-enabled tool to analyze social media. The company has reported that online searches it processed were looking for gang members and related activities.

Social Media Surveillance Tech

Social media scraping, the process of downloading information from social media accounts, including images, posts, likes, and associated accounts (e.g., “Friends”), begins on ethically precarious grounds as primary social media platforms such as Facebook and X (formerly Twitter) explicitly forbid scraping for surveillance purposes in their terms of service.

Link Analysis, a method of inferring group memberships and beliefs, can be a powerful tool when dealing with international criminal groups, but it becomes much more tenuous with tragic consequences when dealing with young people. There is a strong amount of social pressure to comment on or like posts that pass in front of the user, irrespective of how factually remote the relationship might be. This led one Black teen in New York City to be sent to Rikers Island for over a year, and most of it in solitary confinement, in large part because of a vague association with a criminal gang. The district attorney relied on photos of the teen posing with a “crew” (a loose affiliation based on a block or housing development) as well as several “liked” social media posts. There was no evidence of any criminal association. They were merely the teen’s brother, friends, and neighbors. In a sense, police use of social media in this way demands that young people isolate themselves from their neighbors or risk an assumed affiliation.

Facial Recognition Technology (“FRT”) attempts to identify people based on facial photos. The success of FRT depends on several factors, including image quality, lighting, race, and gender, in addition to the algorithm used and the skills and ethics of the technician using it. Still, FRT has exploded in use recently. One company, Clearview AI, is being used by thousands of law enforcement agencies and claims to have sourced many of its images by scraping the internet. Its untested, and therefore unproven, technology can be used by cops to identify potential gang members using probe images pulled from social media despite the fact that even the very best FRT systems as tested by the National Institute of Standards and Technology misidentified people of color at a much greater rate that grew dramatically for “wild” images, such as those drawn from social media.

Keyword searches are simple in concept but are extremely likely to result in an overwhelming number of irrelevant hits that would obscure actual criminal activity. That is what police in Jacksonville, Florida, found as their tool “regularly yielded false alarms,” according to Ben Cornack at Jacksonville.com. When the police department searched for the word “bomb,” they were presented with posts describing pizza and beer as “the bomb.”

Translation tools have their own problems, sometimes giving very puzzling results, especially with regard to African American Vernacular English (“AAVE”). In a 2017 study, one system misidentified AAVE as being Danish and even assigned a 99.9% confidence rating to its finding. Police in Israel detained a man for posting the words “attack them.” Only, what he actually posted in Arabic was “Good morning,” but the automated tool mistranslated it. Anyone who speaks more than one language can attest to the failures of even the most sophisticated translation tools available today as they struggle to translate non-English and vernacular languages.

In order to scrape ever more accounts without a warrant, cops and their tools often turn to fake and undercover accounts. Facebook, as an example, allows its users to identify the visibility of posts. However, if an account can be persuaded to accept a friend request, the new friend may be able to access posts, pictures, and account information about friends and family members, including the ability to exchange and view private messages. Some surveillance software companies will create new accounts in bulk in order to access otherwise private accounts, despite fake account prohibitions by the platforms. Nevertheless, police continue to buy licenses to use the software that is itself violating license agreements.

These activities do not seem to be too different from undercover police work, though there are some significant distinctions. First, in-person undercover officers are unlikely to pass themselves off as childhood friends or relatives. In addition, in-person undercover work tends to be more restrictive. Online undercover work often fails to establish basic limitations, like the presence of a reasonable suspicion of criminal activity or the fake accounts be subjected to supervisory review. The NAACP sued the city of Minneapolis because it was targeting activists. The suit resulted in a consent decree that outlined new oversight and procedural systems with regard to undercover online accounts. While consent decrees have been an important tool for curbing abuses in Minneapolis and New York City, it is a cumbersome tool to apply to every police department in the country.

Best Practice Suggestions

Since some amount of social media monitoring may be reasonable and necessary, the Brennan Center offers a list of best practices to address some of those concerns, which are summarized below:

1)  With regard to criminal investigations, the information gathered should be placed in the investigation files as soon as possible and should only be kept if there is a reasonable suspicion it contains evidence of criminal activity. If it is shared with another agency, that agency should have a memorandum of agreement that also limits how the data will be kept, used, and shared.

2)  Public events require special care when social media is monitored for public safety and “should only be undertaken when specific, articulable, and credible facts demonstrate a public safety concern.” Any such monitoring should be well documented by someone of high rank such as the police chief or a designee. Furthermore, if no indication of criminal activity is found, the social media data should not be maintained.

3)  Social media should be evaluated and authenticated before being used in a criminal investigation.

4)  Undercover accounts should be used sparingly, if at all. Such accounts have a high risk of abuse and an inherent lack of transparency, as well as platform policies preventing their use. These policies would include:

Documentation that there is no less invasive means available and that a warrant or subpoena are impossible or ineffective.

A showing of cause that the use of impersonated/fake accounts will likely obtain information related to a small category of serious crimes. The list of applicable crimes will have been previously defined.

Supervisory approval of the account, officer who will use it, and its purpose.

Documentation of the names of the persons with whom the officer will connect.

Regular, ongoing reviews no more than 45 days apart that confirm its continued necessity.

Automatic termination of account access at the end of a given period unless the chief of police or designee authorizes an extension. In addition, if a specific person is being impersonated, as in a stalker investigation, the permission of the impersonated individual should be acquired. That individual should have the ability to withdraw the permission at any time.

5)  Any social media monitoring efforts must protect against profiling and targeting Constitutionally protected activity. It is to be prohibited if it is in any way based on race, religion, ethnic or national origin, gender or gender identity, sexual orientation or characteristics, or immigration status. Any exercise of First Amendment rights must first be well documented and approved or unavoidably necessary and every reasonable precaution taken.

6)  There must be protections to prevent retaliation against internal whistleblowers who reveal any social media abuses.

7)  Social media monitoring vendors should be required to abide by the foregoing requirements and demonstrate compliance. In the interest of transparency, police departments that engage in social media monitoring should produce regular reports at least every two years.

Finally, the Brennan Center recommends that an independent auditor be appointed to verify compliance with departmental policies and Constitutional safeguards. 

Sources: brennancenter.org, Howard Law Journal

Subscribe today

Already a subscriber? Login