Today, the Colorado Supreme Court became the first state supreme court in the country to address the constitutionality of a keyword warrant—a digital dragnet tool that allows law enforcement to identify everyone who searched the internet for a specific term or phrase. In a weak and ultimately confusing opinion, the court upheld the warrant, finding the police relied on it in good faith. EFF filed two amicus briefs and was heavily involved in the case.

The case is People v. Seymour, which involved a tragic home arson that killed several people. Police didn’t have a suspect, so they used a keyword warrant to ask Google for identifying information on anyone and everyone who searched for variations on the home’s street address in the two weeks prior to the arson.

Like geofence warrants, keyword warrants cast a dragnet that require a provider to search its entire reserve of user data—in this case, queries by one billion Google users. Police generally have no identified suspects; instead, the sole basis for the warrant is the officer’s hunch that the suspect might have searched for something in some way related to the crime.

Keyword warrants rely on the fact that it is virtually impossible to navigate the modern Internet without entering search queries into a search engine like Google's. By some accounts, there are over 1.15 billion websites, and tens of billions of webpages. Google Search processes as many as 100,000 queries every second. Many users have come to rely on search engines to such a degree that they routinely search for the answers to sensitive or unflattering questions that they might never feel comfortable asking a human confidant, even friends, family members, doctors, or clergy. Over the course of months and years, there is little about a user’s life that will not be reflected in their search keywords, from the mundane to the most intimate. The result is a vast record of some of users’ most private and personal thoughts, opinions, and associations.

In the Seymour opinion, the four-justice majority recognized that people have a constitutionally-protected privacy interest in their internet search queries and that these queries impact a person’s free speech rights. The federal Supreme Court has held that warrants like this one that target speech are highly suspect so courts must apply constitutional search-and-seizure requirements with “scrupulous exactitude.” Despite recognizing this directive to engage in careful, in-depth analysis, the Seymour majority’s reasoning was cursory and at points mistaken. For example, although the court found that the Colorado constitution protects users’ privacy interests in their search queries, it held that the Fourth Amendment does not, due to the third party doctrine, because federal courts have held that there is no expectation of privacy in IP addresses. However, this overlooks the queries themselves, which many courts have suggested are more akin to the location information that was found to be protected in Carpenter v. United States. Similarly, the Colorado court neglected to address the constitutionality of Google’s initial search of all its users’ search queries because it found that the things seized—users’ queries and IP addresses—were sufficiently narrow. Finally, the court merely assumed without deciding that the warrant lacked probable cause, a shortcut that allowed the court to overlook the warrant's facial deficiency and therefore uphold it on the “good faith exception.”

If the majority had truly engaged with the deep constitutional issues presented by this keyword warrant, it would have found, as the three-justices dissenting on this point did, that keyword warrants “are tantamount to a high-tech version of the reviled ‘general warrants’ that first gave rise to the protections in the Fourth Amendment.” They lack probable cause because a mere hunch that some unknown person might have searched for a specific phrase related to the crime is insufficient to support a search of everyone’s search queries, let alone a specific, previously unnamed individual. And keyword warrants are insufficiently particular because they do next to nothing to narrow the universe of the search.

We are disappointed in the result in this case. Keyword warrants not only have the potential to implicate innocent people, they allow the government to target people for sensitive search terms like the drug mifepristone, or the names of gender-affirming healthcare providers, or information about psychedelic drugs. Even searches that refer to crimes or acts of terror are not themselves criminal in all or even most cases (otherwise historians, reporters, and crime novelists could all be subject to criminal investigation). Dragnet warrants that target speech have no place in a democracy, and we will continue to challenge them in the courts and to support legislation to ban them entirely.

Google announced this week that it will be making several important changes to the way it handles users’ “Location History” data. These changes would appear to make it much more difficult—if not impossible—for Google to provide mass location data in response to a geofence warrant, a change we’ve been asking Google to implement for years.

Geofence warrants require a provider—almost always Google—to search its entire reserve of user location data to identify all users or devices located within a geographic area during a time period specified by law enforcement. These warrants violate the Fourth Amendment because they are not targeted to a particular individual or device, like a typical warrant for digital communications. The only “evidence” supporting a geofence warrant is that a crime occurred in a particular area, and the perpetrator likely carried a cell phone that shared location data with Google. For this reason, they inevitably sweep up potentially hundreds of people who have no connection to the crime under investigation—and could turn each of those people into a suspect.

Geofence warrants have been possible because Google collects and stores specific user location data (which Google calls “Location History” data) altogether in a massive database called “Sensorvault.” Google reported several years ago that geofence warrants make up 25% of all warrants it receives each year.

Google’s announcement outlined three changes to how it will treat Location History data. First, going forward, this data will be stored, by default, on a user’s device, instead of with Google in the cloud. Second, it will be set by default to delete after three months; currently Google stores the data for at least 18 months. Finally, if users choose to back up their data to the cloud, Google will “automatically encrypt your backed-up data so no one can read it, including Google.”

All of this is fantastic news for users, and we are cautiously optimistic that this will effectively mean the end of geofence warrants. These warrants are dangerous. They threaten privacy and liberty because they not only provide police with sensitive data on individuals, they could turn innocent people into suspects. Further, they have been used during political protests and threaten free speech and our ability to speak anonymously, without fear of government repercussions. For these reasons, EFF has repeatedly challenged geofence warrants in criminal cases and worked with other groups (including tech companies) to push for legislative bans on their use.

However, we are not yet prepared to declare total victory. Google’s collection of users’ location data isn’t limited to just the “Location History” data searched in response to geofence warrants; Google collects additional location information as well. It remains to be seen whether law enforcement will find a way to access these other stores of location data on a mass basis in the future. Also, none of Google’s changes will prevent law enforcement from issuing targeted warrants for individual users’ location data—outside of Location History—if police have probable cause to support such a search.

But for now, at least, we’ll take this as a win. It’s very welcome news for technology users as we usher in the end of 2023.