Did you know that anything created by federal government employees, under American law, can’t be copyrighted? They go right into the public domain. That means it’s impossible for there to be any copyright infringement on, say, a report by the Special Counsel on the investigation into the President and his presidential campaign. But the copyright-enforcement bots on Scribd have been busy taking down copies of the Mueller report all the same.

Scribd is a service that allows users to upload documents for easy embedding. It’s commonly used by journalists to attach things like legal complaints or court records to articles based on those records. It is therefore not at all surprising that a number of people and news outlets uploaded the Mueller report to Scribd.

Scribd has an automated filter that searches uploads to check for alleged copyright infringement called BookID. BookID, like YouTube’s Content ID, has a propensity for false positives. This apparently happens so often that Scribd’s own page on the system has a whole section devoted to “false positives,” explaining:

The BookID database may contain reference samples from educational textbooks and other works that contain long excerpts of classic literature, religious texts, legal documents, and government publications that are typically in the public domain. This can occasionally result in the removal of uncopyrighted, authorized, or public domain material from Scribd.

False positives also happen enough that Scribd sent a letter—obtained by Quartz—to people whose uploads of the Mueller report were taken down, to lower people’s expectations. The letter explains that a) all automated systems will flag legitimate content b) the volume of content on Scribd means no one checks on the matches before the alleged infringing content is taken down and c) sometimes legitimate uploads are taken down just for being duplicates.

Note again: it is impossible for a copy of the Mueller report to be infringement since it cannot be copyrighted. Filters like BookID don’t work, and in this case, actively work against the public interest. As Quartz points out, the European Union recently passed a new copyright directive that will require platforms to take proactive measures against possible infringement. In practice, that means filters like BookID. As instances like this and all the other similar ones EFF has been documenting for years prove, filters harm civil discourse, and requiring a system like the one Scribd itself points out has huge faults is a mistake.