Leveraging close ties to Twitter, controversial artificial intelligence startup Dataminr helped law enforcement digitally monitor the protests that swept the country following the killing of George Floyd, tipping off police to social media posts with the latest whereabouts and actions of demonstrators, according to documents reviewed by The Intercept and a source with direct knowledge of the matter.
The monitoring seems at odds with claims from both Twitter and Dataminr that neither company would engage in or facilitate domestic surveillance following a string of 2016 controversies. Twitter, up until recently a longtime investor in Dataminr alongside the CIA, provides the company with full access to a content stream known as the “firehose” — a rare privilege among tech firms and one that lets Dataminr, recently valued at over $1.8 billion, scan every public tweet as soon as its author hits send. Both companies denied that the protest monitoring meets the definition of surveillance.
Dataminr helps newsrooms, corporations, and governments around the world track crises with superhuman speed as they unfold across social media and the wider web. Through a combination of people and software, the company alerts organizations to chatter around global crises — wars, shootings, riots, disasters, and so forth — so that they’ll have a competitive edge as news is breaking. But the meaning of that competitive edge, the supercharged ability to filter out important events from the noise of hundreds of millions of tweets and posts across social media, will vary drastically based on the customer; the agenda of a newspaper using Dataminr to inform its breaking news coverage won’t be the same as the agendas of a bank or the FBI. It’s this latter category of Dataminr’s business, lucrative government work, that’s had the firm on the defensive in recent years.
In 2016, Twitter was forced to reckon with multiple reports that its platform was being used to enable domestic surveillance, including a Wall Street Journal report on Dataminr’s collaboration with American spy agencies in May; an American Civil Liberties Union report on Geofeedia, a Dataminr competitor, in October; and another ACLU investigation into Dataminr’s federal police surveillance work in December. The company sought to assure the public that attempts to monitor its users for purposes of surveillance were strictly forbidden under its rules, and that any violators would be kicked off the platform. For example, then-VP Chris Moody wrote in a company blog post that “using Twitter’s Public APIs or data products to track or profile protesters and activists is absolutely unacceptable and prohibited.” In a letter to the ACLU, Twitter public policy chief Colin Crowell similarly wrote that “the use of Twitter data for surveillance is strictly prohibited” and that “Datatminr’s product does not provide any government customers with … any form of surveillance.”
Twitter also said that Dataminr, one of its “official partners,” would “no longer support direct access by fusion centers” to information such as tweet locations; fusion centers are controversial facilities dedicated to sharing intelligence between the federal government and local police. Dataminr at the same time announced it would no longer provide a product for conducting geospatial analysis “to those supporting first reponse” and added that such clients did not have “direct firehose access.”
But based on interviews, public records requests, and company documents reviewed by The Intercept, Dataminr continues to enable what is essentially surveillance by U.S. law enforcement entities, contradicting its earlier assurances to the contrary, even if it remains within some of the narrow technical boundaries it outlined four years ago, like not providing direct firehose access, tweet geolocations, or certain access to fusion centers.
Dataminr relayed tweets and other social media content about the George Floyd and Black Lives Matter protests directly to police, apparently across the country. In so doing, it used to great effect its privileged access to Twitter data — despite current terms of service that explicitly bar software developers “from tracking, alerting, or monitoring sensitive events (such as protests, rallies, or community organizing meetings)” via Twitter.
And despite Dataminr’s claims that its law enforcement service merely “delivers breaking news alerts on emergency events, such as natural disasters, fires, explosions and shootings,” as a company spokesperson told The Intercept for a previous report, the company has facilitated the surveillance of recent protests, including nonviolent activity, siphoning vast amounts of social media data from across the web and converting it into tidy police intelligence packages.
Dataminr’s Black Lives Matter protest surveillance included persistent monitoring of social media to tip off police to the locations and activities of protests, developments within specific rallies, as well as instances of alleged “looting” and other property damage. According to the source with direct knowledge of Dataminr’s protest monitoring, the company and Twitter’s past claims that they don’t condone or enable surveillance are “bullshit,” relying on a deliberately narrowed definition. “It’s true Dataminr doesn’t specifically track protesters and activists individually, but at the request of the police they are tracking protests, and therefore protesters,” this source explained.
“At the request of the police they are tracking protests, and therefore protesters.”
According to internal materials reviewed by The Intercept, Dataminr meticulously tracked not only ongoing protests, but kept comprehensive records of upcoming anti-police violence rallies in cities across the country to help its staff organize their monitoring efforts, including events’ expected time and starting location within those cities. A protest schedule seen by The Intercept shows Dataminr was explicitly surveilling dozens of protests big and small, from Detroit and Brooklyn to York, Pennsylvania, and Hampton Roads, Virginia.
Company documents also show the firm instructed members of its staff to look for instances of “lethal force used against protesters by police or vice-versa,” “property damage,” “widespread arson or looting against government or commercial infrastructure,” “new instances of officer-involved shootings or death with potential interpretation of racial bias,” and occasions when a “violent protests spreads to new major American city.” Staff were also specifically monitoring social media for posts about “Officers involved in Floyd’s death” — all of which would be forwarded to Dataminr’s governmental customers through a service named “First Alert.”
The Dataminr documents on protest monitoring seen by The Intercept do not specify if they are used for news clients, police clients, or both. But a Dataminr document from October 2019 listed within the company’s “law enforcement footprint” the New York Police Department, Los Angeles Police Department, Chicago Police Department, and Louisiana State Police. The LAPD told The Intercept it conducted a trial of Dataminr but chose not to enter a contract and did not use the system in connection with BLM protests. The Louisiana State Police declined to comment, citing a state secrecy law. NYPD did not comment and CPD could not be reached for comment. In January 2019, a New York court ordered the NYPD to turn over records about its use of Dataminr resulting from a New York Civil Liberties Union lawsuit over alleged surveillance of Black Lives Matter activists.
“Dataminr is providing information for local police, including [many] metropolitan police departments in cities facing protests,” the source said. “They are some of Dataminr’s biggest clients and they set the agenda.” Dataminr spokesperson Kerry McGee declined to comment on the company’s clientele.
And Dataminr alert emails sent to the Minneapolis Police Department, obtained via a public records request, show the company collected, bundled, and captioned Twitter content relevant to the anti-police brutality protests and forwarded it directly to police as these events unfolded, including information on apparently nonviolent protests. The emails show Dataminr relaying the locations and images of Black Lives Matter protesters in the city where George Floyd lived and was killed, and where the nationwide wave of outrage against police abuse was launched, a fact difficult to square with the company’s claim that it doesn’t provide its governmental customers with “any form of surveillance.” The location information in the alerts underline that while Dataminr may not technically have direct access to the geolocational data attached to many tweets by Twitter, the texts and images of the tweets relayed to the police often contain overt geographical references, or have such references added manually by Dataminr staff.
While some of the alerts are sourced from the tweets of local and national news reporters, many are attributed to the accounts of ordinary bystanders — what the system calls “eyewitnesses” — who were either watching or attending the rallies and tweeting in a completely personal capacity. In one First Alert message relayed to the MPD on May 31, six days after Floyd’s murder, Dataminr alerted police to a tweet reading “peaceful protest outside US Bank Stadium in downtown Minneapolis. End racism. End police brutality. End inequality and inequities. #JusticeForFloyd #Minneapolisprotest #BlackLivesMatters,” along with a photo snapped by the tweeter. The accompanying caption, provided by Dataminr’s human staff, specified that this group of protesters had been “seen at US Bank Stadium on 400 block of Chicago Avenue.” Another First Alert notification sent to the MPD three days prior tipped off police to this supposed public safety threat: “Protesters seen sitting on street in front of security officers in Oakdale, MN.” Another monitored tweet and accompanying photo relayed to MPD by Dataminr reads, simply, “Peaceful protest at Lake & Lyndale.”
A tweet and photo relayed to Minneapolis police reads, simply, “Peaceful protest at Lake & Lyndale.”
First Alert also scans other popular platforms like Snapchat and Facebook, the latter being particularly useful for protest organizers trying to rapidly mobilize their communities. On at least one occasion, according to MPD records, Dataminr was able to point police to a protest’s Facebook event page before it had begun.
Some of Dataminr’s alerts passed along dubious information. For example, on May 28, the company passed along a discredited claim about billionaire philanthropist George Soros, informing the MPD that “Commentator Candace Owens claims Minneapolis, MN chief of police says many protesters are not from the city and claims investor George Soros is funding protesters through Open Society Foundation.”
This apparently glaring contradiction by Dataminr, still publicly claiming it would never engage in surveillance while simultaneously facilitating the surveillance of protests, hasn’t been lost on the company’s staff. At a virtual staff meeting in June, a recording of which was obtained by The Intercept, a Dataminr manager attempted to explain why the company’s persistent monitoring of First Amendment activity on behalf of the police was not, in fact, surveillance. The manager, identified by the source as executive vice president Jason Wilcox, granted that there were likely Dataminr staff pondering some difficult questions: “How does our technology, how does our company, how does our platform, play in these types of unfolding events that are out there?” — an allusion to the nationwide protests that were by then in their heady first week. “We sell to law enforcement. What does that mean?” Wilcox’s defense of Dataminr was based mostly on a sort of linguistic distinction: that relaying data to the police isn’t a form of surveillance, but a form of ideologically neutral newsgathering. In an alternate euphemism, Wilcox described the surveillance alerts forwarded to police as “situational awareness through real time events, [in] many of which people’s lives are at stake, and they can respond more quickly and save lives.”
This is generally the same reason Twitter and Dataminr’s PR teams describe this governmental product as a source of “news alerts,” not intelligence — a rationale that largely obscures the major differences between what, say, a newspaper might do with rapidly updated information about a protest against policing versus what the police might want to do with that same data.
Wilcox added that Dataminr’s protest surveillance, far from presenting any chilling effect on political expression or free assembly, was a force of progressivism and reform: “We alert on events where members of law enforcement overstep their bounds,” Wilcox claimed. “We found abuse of power. … Ultimately what we’re doing is we’re providing a check and balance for [police]. … Those alerts provide context to the world keeping people safe, and enabling people to do so in a way that isn’t about trying to invade user privacy, but quite the opposite. It helps magnify their voice.”
Asked about the comments, Dataminr’s McGee wrote, “Dataminr does not comment on internal company meetings.” Wilcox did not respond to a request for comment.
Wilcox also defended Dataminr’s work with police by emphasizing the firm’s close ties to Twitter, the great firehose benefactor, which, according to Wilcox, “is often one of the first social media platforms to reach out and protect privacy, they seem to be most attuned to it, they’re very concerned with ensuring that their platform is not misused.”
“Those alerts provide context to the world, keeping people safe.”
Dataminr’s internal justification of its work for police also rested in part on the argument that it’s not as nefarious as it could be: “We look at lots of different companies leveraging social media, and they have often, not everyone, but often, a very different set of goals,” Wilcox explained. “Their goal is to help with surveillance. They build users graphs, they track users as they go across different social media platforms, they follow what a person says over time. And we do not do that.” Wilcox named a few other mechanisms he said showed how he’d “worked hard to ensure that our technology cannot be casually misused here,” namely built-in limits on what keywords police can use to tailor their “news alerts.”
But according to the source with direct knowledge of Dataminr’s protest monitoring, this is misleading: There’s nothing built into First Alert that would prevent police from filtering or manually searching the intelligence they receive from Dataminr for specific terms, such as “#BLM” or “antifa.” Once a protest tweet is run through Dataminr’s system and spit out the other end into a police department’s inboxes, in other words, Dataminr loses control over how the information is used. This image of technological restraint also differs considerably from the pitch Dataminr gives police. An apparent 2019 Dataminr slide deck from a company presentation to the FBI, included in a recent online data dump known as “BlueLeaks,” stated that “Dataminr’s mission is to integrate all publicly available data signals to create the dominant information discovery platform,” and touted a client’s ability to customize “user-defined criteria” for alerts like “topic selection” and “geographic filters.” The end goal: “Reduce the time between an event and client action.”
When asked about Dataminr’s work with law enforcement as outlined above, both Twitter and Dataminr adopted a similar defense: This isn’t surveillance because we have a policy against surveillance, which therefore means we don’t engage in surveillance. Neither firm would comment or discuss how exactly the above does not meet the definition of surveillance, nor would they provide the institutional definitions of such as defined by either company.
“We see a societal benefit in public Twitter data being used for news alerting, first responder support, and disaster relief,” said Twitter spokesperson Lindsay McCallum, who added that Dataminr’s First Alert tool “is in compliance with our developer policy” banning surveillance. “First Alert is not permitted to be used for surveillance of any kind by First Alert users,” Dataminr’s McGee told The Intercept. In response to a screenshot copy of the tweet Dataminr forwarded to Minneapolis police regarding the exact location of a group of protesters, McGee claimed that this was flagged for the department because it showed traffic problems, not protesters. “Alerts on an intersection being blocked are news alerts, not monitoring protests or surveillance,” said McGee. “A local news organization would also cover major intersections being blocked as a news story — this is not surveillance.”
But to some surveillance scholars, legal experts, and activists, there’s little doubt about what Dataminr is up to, and what Twitter is enabling, no matter what careful terminology they use. According to Brandi Collins-Dexter, a campaign director with the civil rights group Color of Change, Dataminr’s practices are an example of “if it walks like a duck and talks like a duck,” with regards to surveillance. “We know that law enforcement agencies spend a breathtaking amount of money to aggressively track, target, and surveil Black communities,” said Collins-Dexter. “Twitter can’t have it both ways, courting Black activists and marketing themselves as the pre-eminent tool for organizing against injustice, while turning a blind eye to the number of companies that are contracting with them for the clear intent of surveillance.”
“Twitter can’t have it both ways, courting Black activists and marketing themselves as the pre-eminent tool for organizing against injustice, while turning a blind eye to companies that are contracting with them for the clear intent of surveillance.”
Steven Renderos, the executive director of civil rights group MediaJustice, echoed this sentiment. “It’s troubling that that Dataminr is providing services to police and it’s flawed logic to think there’s no harm in turning over Twitter posts to cops,” said Renderos. “The police have a history of using social media to track Black activists. Dataminr’s practices is just the latest example of how tech companies are fueling racist policing in the United States.”
“If Dataminr is sharing posts about demonstrations and protesters with police, that would be incredibly concerning and it would be difficult to understand how that practice doesn’t facilitate police surveillance in violation of Twitter’s own policies,” said Matt Cagle, an attorney with the ACLU of Northern California. “Social networks like Twitter need to protect users and ensure that developers are not sharing their First Amendment expression with law enforcement agencies, a practice that potentially exposes people — particularly Black, Indigenous, and people of color — to further surveillance and state violence.”
Andrew Ferguson, a visiting law professor at American University, rejected the companies’ contention that because Dataminr only ingests public tweets, the system is only capable of news gathering — as if police snapping pictures of demonstrators would be better understood as photojournalism, not photo surveillance. “Monitoring activities and forwarding information to police is clearly surveillance,” explained Ferguson, author of “The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement.” “If the police were a data-based advertising company we would say this was consumer surveillance. If the police were tracking protestors directly we would call it government surveillance. A forwarding of the same information and calling it ‘news’ is still surveillance.”
Whether Twitter’s hundreds of millions of users will buy the argument that automatically relaying tweets to the police is mere innocent newsgathering remains an open question, but for most of them a moot one: Outside of laborious public records requests, it’s hard to imagine how someone could learn if their protest tweets were swallowed into the algorithms by a government contractor. Or, one could think of it the way Jason Wilcox urged his staff: “All those voices, where we get to amplify that for everybody. … It’s pretty impressive. It’s an amazing event.”