Torrenting court order: why your takedown records matter

A US magistrate judge has ordered a major US technology company to name the employees behind its decisions to download pirated book archives by torrent. No liability has been found.

On 20 October 2026, a piracy news outlet reported that a US magistrate judge has ordered the company to identify the people involved in its decisions to obtain pirated book collections by torrent. The order came in a case brought by a US business magazine publisher, which runs alongside the better known class action by authors including a novelist and a comedian. It is a discovery order in live litigation. The company has not been found to have infringed anything, and nobody named in response to it may be characterised as a wrongdoer.

For a publisher, studio or rights holder's legal team, the interest is not in the company. It is in what the order reveals about how hard it is to establish who decided to take content and when, even inside a company with full records, and in what that means for the records a rights holder keeps on its own side.

What happened

The underlying facts, as the outlet reports them, are not in dispute in the way the law is. The company does not deny that it used torrenting to download a collection from one large shadow library and data from another, as training material for its AI models. In the authors' class action, the court has already ruled that training on the books was fair use. What remains live in that case, and what the publisher's separate suit turns on, is the torrenting part: whether, by using a protocol that uploads as it downloads, the company distributed pirated works to others.

The publisher had asked the company to identify everyone who "made, authorized, or approved" the decisions to obtain the datasets by torrent. The company's position, as reported, was that it could not identify anyone who "formally" made or approved such decisions, that decisions were often reached by team consensus, that the people who ran the downloads were making technical choices, and that it knew of no one who approved uploading or distributing the data. The judge found that the question was not limited to formal decisions, that the company's answers mostly addressed how the data was used rather than how it was obtained, and that collaborative decision making does not excuse a company from naming everyone involved. No compliance deadline is reported.

This article does not name any employee of the company, whether or not the source does. They are participants in a discovery dispute, not defendants, and nothing has been found against any of them.

What the news actually is

It would be easy to read this as a court closing in on the company. It is not that, or not yet. Fair use for training has already gone the company's way in the authors' case. Whether the torrenting itself was fair use, and whether the company in fact distributed anything to anyone, are both undecided.

What the news actually is, for anyone outside the case, is a demonstration of how the acquisition step has become discoverable. For most of the history of online infringement, the question of who inside an organisation decided to take a copy, and when, and whether they knew, has been unanswerable from the outside. This order shows a court treating it as answerable, and treating "it was a team decision" as a reason to name the team rather than a reason to name nobody.

The mechanism, not the outrage

Infringement cases turn on facts that are usually held by the other side. Who downloaded, from where, when, what settings were used, who signed off, what the lawyers said. A rights holder bringing a claim typically knows only that its content ended up somewhere it should not have. Everything else has to be prised out through discovery, which is slow, expensive and contested, as this case shows.

The command histories ordered in September and the names ordered in October are that prising. They matter because they turn a general allegation into a specific sequence: this person, in this discussion, on this date, chose this. That is what willfulness arguments and contributory infringement arguments are built from, and the publisher said as much, telling the court the answers "bear on" both.

Here is the part that transfers to any rights holder. The side with the better dated record wins the argument about what happened. The company has the records, which is why they can be ordered. Most rights holders, on their own side of the ledger, do not. They know their content was taken but cannot say, with dates and evidence, when it first appeared, where, how many times, and what they did about it each time. For what building that record costs in practice, our breakdown of what takedowns cost is a useful reference.

The gap the reader cannot see

There is a second gap here, and it is one Ceartas should be honest about. The harm in this case is acquisition for AI training. Content was downloaded to be learned from. The harm Ceartas acts on every day is distribution: content put where the public can see or take it. The two are adjacent, and the torrent protocol is exactly where they meet, because downloading by torrent can mean uploading to others at the same time. But they are not the same thing, and nothing a removal service does would have prevented a company from downloading a shadow library onto its own servers.

What a removal operation does produce, as a by product, is the dated record this case shows courts care about. Every time an infringing copy is found, reported and removed, there is a timestamp, a location, a request and a response. Over months that becomes a documented trail of what was taken, where it reappeared, and how the hosts and platforms responded.

Nobody outside litigation gets that. What a rights holder can get, by itself and starting today, is the mirror image: a complete dated record of its own side.

What this does and does not change

This is a discovery order in live litigation. As of 20 October 2026:

• The company has not been found liable for anything, in this case or in the authors' class action, where training was ruled fair use.

• The order decides nothing about whether the torrenting was fair use or whether the company distributed any work.

• It names no wrongdoer, and the people the company must identify are witnesses to a decision, not defendants.

• It creates no new obligation on any AI developer, platform or company outside this case.

• It sets no reported deadline and may itself be contested.

• It concerns acquisition for AI training, not public distribution, and does not directly change how infringing copies online are found or removed.

What it does change is expectations about the acquisition record. A company that acquires content by torrent should now expect to be asked, in court, who decided and when, and should expect "it was a team" to be met with "name the team". The wider push to put obligations on intermediaries, including the US site blocking bills, sits alongside this as a second front: one about who distributes, the other about who takes.

What a rights holder can do now

1. Start a dated register of appearances. For each work, record every unauthorised appearance you find: URL, host, date found, how found. This is the record a court will eventually ask you for, and it cannot be reconstructed later.

2. Record every request and every response. Date sent, to whom, what was asked, what came back, when the content came down or did not. The gap between request and removal is evidence in itself.

3. Separate delisting from removal in your records. A copy that disappeared from search results but stayed on the host was not removed. Log the two outcomes differently, because they mean different things for damages and for willfulness. Our guide to filing a copyright complaint with Google covers the delisting route; the host request is the removal.

4. Preserve what you find. Capture the infringing page or file as it appeared, with the date. Content that has since been removed or changed is hard to prove without a contemporaneous copy.

5. Keep the record even when you are not litigating. Most rights holders will never sue the party that took their content. The ones who do will need years of record, and will be glad they started in the years they were not planning to.

How Ceartas approaches it

We do not act on AI training acquisition, and this article has said so. What we do produces the record this case shows matters.

We monitor for a client's content across platforms, sites and file hosts, and every finding is logged with its location and date. The focus is removal at the host rather than visibility in search, and each request and each response is recorded, so the client holds a dated trail of what was taken, where it reappeared, what was asked of whom and what happened. Removal is ongoing, because the same content returns through a new host or URL, and each return is logged and reported again. If the client's lawyers one day need to show a court what happened and when, the record already exists.

Build the record before you need it

The publisher's case may take years to reach a conclusion, and most rights holders will never be in that courtroom. The copies of your work appearing online this week do not have to wait for one. If you want infringing copies found, removed at the host and logged, with a dated record of every appearance, request and response, see how Ceartas automates monitoring, removal and evidence at scale.

More resources:

• True cost of a DMCA takedown: 2025 complete breakdown

• Two new US site blocking bills could shift online piracy enforcement

• Google copyright complaint: filing guide