Dataminr, an “official partner” of Twitter, alerted a federal law enforcement agency to pro-abortion protests and rallies in the wake of the reversal of Roe v. Wade, according to documents obtained by The Intercept through a Freedom of Information Act request.
Internal emails show that the U.S. Marshals Service received regular alerts from Dataminr, a company that persistently monitors social media for corporate and government clients, about the precise time and location of both ongoing and planned abortion rights demonstrations. The emails show that Dataminr flagged the social media posts of protest organizers, participants, and bystanders, and leveraged Dataminr’s privileged access to the so-called firehose of unrestricted Twitter data to monitor constitutionally protected speech.
“This is a technique that’s ripe for abuse, but it’s not subject to either legislative or judicial oversight,” said Jennifer Granick, an attorney with the American Civil Liberties Union’s Speech, Privacy, and Technology Project.
The data collection alone, however, can have a deleterious effect on free speech. Mary Pat Dwyer, the academic program director of the Institute for Technology Law and Policy at Georgetown University, told The Intercept, “The more it’s made public that law enforcement is gathering up this info broadly about U.S. residents and citizens, it has a chilling effect on whether people are willing to express themselves and attend protests and plan protests.”
The documents obtained by The Intercept are from April to July 2022, during a period of seismic news from the Supreme Court. Following the leak of a draft decision that the court would overturn Roe v. Wade, the cornerstone of reproductive rights in the U.S., pro-abortion advocates staged massive protests and rallies across the country. This was not the first time Dataminr helped law enforcement agencies monitor mass demonstrations in the wake of political outcry: In 2020, The Intercept reported that the company had surveilled Black Lives Matter protests for the Minneapolis Police Department following the murder of George Floyd.
The Marshals Service’s social media surveillance ingested Roe-related posts nearly as soon as they began to appear. In a typical alert, a Dataminr analyst wrote a caption summarizing the social media data in question, with a link to the original post. On May 3, 2022, the day after Politico’s explosive report on the draft decision, New York-based artist Alex Remnick tweeted about a protest planned later that day in Foley Square, a small park in downtown Manhattan surrounded by local and federal government buildings. Dataminr quickly forwarded their tweet to the Marshals. That evening, Dataminr continued to relay information about the Foley Square rally, now in full swing, with alerts like “protestors block nearby streets near Foley Square,” as well as photos of demonstrators, all gleaned from Twitter.
The following week, Dataminr alerted the Marshals when pro-abortion demonstrators assembled at the Basilica of St. Patrick’s Old Cathedral in Manhattan, coinciding with a regular anti-abortion event held by the church. Between 9:06 and 9:53 that morning, the Marshals received five separate updates on the St. Patrick’s protest, including an estimated number of attendees, again based on the posts of unwitting Twitter users.
In the weeks and months that followed, the emails show that Dataminr tipped off the Marshals to dozens of protests, including many pro-abortion gatherings, from Maine to Wisconsin to Virginia, both before and during the demonstrations. Untold other protests, rallies, and exercises of the First Amendment may have been monitored by the company; in response to The Intercept’s public records request, the Marshals Service identified nearly 5,000 pages of relevant documents but only shared about 800 pages. The U.S. Marshals Service did not respond to a request for comment.
The documents obtained by The Intercept are email digests of social media activity that triggered alerts based on requested search terms, which appear at the bottom of the reports. The subscribed topics have ambiguous names like “SCOTUS Mentions,” “Federal Courthouses and Personnel Hazards_V2,” “Public Safety Critical Events,” “Attorneys,” and “Officials.” The lists suggest that the Marshals were not specifically seeking information on abortion rallies; rather, the agency had cast such a broad surveillance net that large volumes of innocuous First Amendment-protected activity regularly got swept up as potential security threats. What the Marshals did with the information Dataminr collected remains unknown.
“The breadth of these search categories and terms is definitely going to loop in political speech. It’s a certainty,” Granick told The Intercept. “It’s a reckless indifference to the fact that you’re going to end up spying on core constitutionally protected political activity.”
The oldest law enforcement agency in the U.S., the Marshals are a niche holdover of early American policing, immortalized in cowboy movies and tales of the Wild West. Today, the Marshals Service retains a unique mission among federal agencies, consisting largely of transporting prisoners, hunting fugitives, and ensuring the safety of federal courts and judicial staff.
While some of the Dataminr alerts aligned with this mission, such as informing the Marshals of protests near courthouses or judges’ homes, others monitored protests in locations without any ostensible relation to the judiciary. The Basilica of St. Patrick’s Old Cathedral is well over a mile from the nearest courthouse and surrounded by trendy cafes and boutiques. Brooklyn’s Barclays Center, a sports and performance venue where a protest organized on Facebook was flagged by Dataminr on May 3, 2022, is nearly a mile from the closest courthouse.
The Marshals’ broad use of social media surveillance is not the first instance of its apparent mission creep in recent years: In 2021, The Intercept reported that a drone operated by the Marshals had spied on Black Lives Matter protests in Washington, D.C.
As an attorney who frequents courthouses, including during protests, Granick rejected the notion that a political rally is a security threat by dint of its proximity to a judiciary building.
“I would say that a tiny, tiny, tiny fraction of protests at courthouses pose any kind of risk of either property damage or personal injury,” she said. “And there’s really no reason to gather information on who is going to that protest, or what their other political views are, or how they’re communicating with other people who also believe in that cause.”
Dataminr sent a regular volley of alerts about planned and ongoing protests at or near the homes of conservative Supreme Court Justices Clarence Thomas, Brett Kavanaugh, and Amy Coney Barrett. On June 24, 2022, Dataminr sent the Marshals an alert that read, “Protest planned for 18:30 at CVS on 5700 Burke Centre Parkway in Burke, VA to travel to residence of US Supreme Court Justice Thomas.” Follow-up alerts noted the protesters were “at entrance to subdivision of neighborhood where US Supreme Court Justice Thomas lives.” A third alert included that the Marshals were already at the protest; it’s unclear why the agency would need to monitor discussion of an event where its marshals were already present.
Only a small fraction of the alerts reviewed by The Intercept include content that could plausibly be construed as threatening, and even those seem to lack any specificity that would make them useful to a federal agency. On May 3, 2022, Dataminr flagged a tweet that read “WE’RE COMING FOR YOU PLANNED PARENTHOOD.” A week later, another tweet exhorted followers to “[b]urn down anti abortion orgs, kick in extremist churches and smash the homes of the oppressors.”
“There’s an assumption underlying this that someone who complains on Twitter is more dangerous than someone who doesn’t complain on Twitter.”
The following month, Dataminr reported two tweets to the Marshals that appeared to be more hyperbolic fantasies than credible threats. One user tweeted that they would pay to watch the Supreme Court justices who overturned Roe burn alive, while another cited an individual who tweeted, “I’m not not advocating for burning down buildings. But trauma and destruction is kind of the thing that I love.”
At other times, Dataminr seemed incapable of distinguishing between slang and violence. Among several tweets about the 2022 Met Gala inexplicably flagged by Dataminr, the Marshals Service was alerted to a fan account of the actor Timothée Chalamet that tweeted, “i would destroy the met gala” — an online colloquialism for something akin to stealing the show.
These alerts show that despite the claims in its marketing materials, Dataminr isn’t necessarily in the business of public safety, but rather bulk, automated scrutiny. Given the generally incendiary, keyed-up nature of social media speech, a vast number of people might potentially be treated with suspicion by police in the total absence of a criminal act.
“There’s an assumption underlying this that someone who complains on Twitter is more dangerous than someone who doesn’t complain on Twitter,” Granick said. “Inevitably, you have people making decisions about what anger is legitimate and what anger is not.”
Aside from alerts about protests near judges’ homes or courthouses, many of the Dataminr notices appear to have no relevance to American law enforcement. Emails reviewed by The Intercept show that Dataminr alerted the Marshals to social media chatter about Saudi airstrikes in Yemen, attacks in Syria using improvised explosive devices, and political protests in Argentina.
Dataminr represents itself as a “real-time AI platform,” but company sources have previously told The Intercept that this is largely a marketing feint and that human analysts conduct the bulk of platform surveillance, scouring the web for posts they think their clients want to see.
Nonetheless, Dataminr is armed with one technological advantage: the Twitter firehose. For companies willing to pay for it, Twitter’s firehose program provides unfettered access to the entirety of the social network and the ability to automatically comb every tweet, topic, and photo in real time.
The Marshals Service emails also show the extent to which Dataminr is drinking from far more than the Twitter firehose. The emails indicate that the agency is notified when internet users merely mention certain political figures, namely judges and state attorneys general, on Telegram channels or in the comments of news articles.
Although most of the Dataminr alerts don’t include the text of the original posts, those that do often flag innocuous content across the political spectrum, including hundreds of mundane comments from blogs and news websites. In July, for instance, Dataminr reported to the Marshals web comments calling New York Attorney General Letitia James a “racist;” a user saying, “God Bless Gov. Youngkin,” referring to the Virginia governor; and another comment arguing that “Trump wants to hide out in the Oval Office from the responsibility and any accountability for what he did on January 6th and before.” When Ohio Attorney General Dave Yost made national headlines after suggesting that reports of a 10-year-old rape victim denied an abortion may have been fabricated, the Marshals received dozens of alerts about blog comments debating his words.
In some cases, Dataminr appeared incapable of differentiating between people with the same name. On May 18, the Marshals received an alert that “New Jersey District Court Magistrate Judge Jessica S. Allen” was mentioned in a Telegram channel used to organize an anti-Covid lockdown rally in Australia. The text in question appears to be automated, semicoherent spam: “I’ve been a victim of scam, was scared of getting scammed again, but somehow I managed to squeeze out some couple of dollars and I invested with Jessica Allen, damn to my surprise I got my profit within 2 hours.”
Even those sharing links to articles without any added commentary on Telegram fell under Dataminr scrutiny. When one Telegram user shared a July 4, 2022, story from The Hill about Kentucky Attorney General Daniel Cameron’s request that the Supreme Court put the state’s abortion ban back in place, it was flagged to the U.S. Marshals within an hour.
“Discussions of how people view political officials governing them, discussions of constitutional rights, planning protests — that’s supposed to be the most protected speech,” Georgetown’s Dwyer said. “And here you have it being swept up and provided to law enforcement.”
At the time the Marshals received the alerts obtained by The Intercept, Dataminr was listed as an “official partner” on Twitter’s website. Since Elon Musk acquired Twitter in October 2022, the company’s partnership with the social media site has continued. Despite his fury against people who might track the location of his private jet, Musk does not appear to have similar misgivings about furnishing federal police with the precise real-time locations of peaceful protesters.
Twitter’s longtime policy forbids third parties from “conducting or providing surveillance or gathering intelligence” or “monitoring sensitive events (including but not limited to protests, rallies, or community organizing meetings).” When asked how Dataminr’s surveillance of protests using Twitter could be compatible with the policy banning the surveillance of protests, Dataminr spokesperson Georgia Walker said in a statement:
Dataminr supports all public sector clients with a product called First Alert which was specifically developed with input from Twitter, and fully complies with Twitter’s policies and the policies of all our data providers. First Alert delivers breaking news alerts enabling first responders to respond more quickly to public safety emergencies. First Alert is not permitted to be used for surveillance of any kind by First Alert users. First Alert provides a public good while ensuring maximum protections for privacy and civil liberties.
Both Twitter, which no longer has a communications team in the Musk era, and Dataminr have denied that the persistent real-time monitoring of the platform on behalf of police constitutes “surveillance” because the posts are public. Civil libertarians and scholars of state surveillance generally reject their argument, noting that other forms of surveillance routinely occur in public spaces — security cameras pointed at the sidewalk, for instance — and that Dataminr is surfacing posts that would likely be hard for police to find through a manual search.
“There is a world of difference between reading through some public tweets and having a service which indexes, stores, aggregates, and makes that information searchable.”
“There is a world of difference between reading through some public tweets and having a service which indexes, stores, aggregates, and makes that information searchable,” Granick said. As is typical with surveillance tools, police are inclined to use Dataminr not necessarily because it’s effective in thwarting or solving crimes, she said, but because it’s easy and relatively cheap. Receiving a constant flow of alerts from Dataminr creates the appearance of intelligence-gathering without any clear objective or actual intelligence.
In the absence of automated tools like Dataminr, police would have to make choices about how to use their finite time to sift through the vastness of social media platforms, which would likely result in more focus on actual criminality instead of harmless political chatter.
“What this technology does is it liberates law enforcement from having to make that economic calculation and enables them to do both,” Granick explained. “And then once the technology does that, in the absence of any kind of regulation, there’s insufficient disincentive to stop them from doing it.”
Following January 6, 2021, lawmakers questioned why police were blindsided by the storming of the U.S. Capitol even though it was openly planned online. There were calls to bolster the government’s ability to monitor social media, which were again sounded in the wake of the recent leak of classified intelligence documents on Discord. These calls, however, ignore the vast scale of social media surveillance already taking place, surveillance that has failed to stop both apparent blows to state security.
While Dataminr and its many competitors stand to profit immensely from more government agencies buying these tools, they have little to say about how they’ll avoid generating even more noise in search of signal.
“Collecting more hay,” Granick said, “doesn’t help you find the needle.”
Correction: May 16, 2023
This story has been updated to use Alex Remnick’s correct pronoun.