People building the future of the social web — interoperable and decentralized — need to protect themselves against copyright liability. Like anyone who creates and operates platforms for user-uploaded content, the hosts of the decentralized social web can take preventive measures to reduce their legal exposure when a user posts material that violates someone’s copyright.

This post gives an overview of the steps to take. It’s meant for operators of Mastodon and other ActivityPub servers, Bluesky hosts, RSS mirrors, and other decentralized social media protocols, and developers of apps for those protocols — but it will apply to other hosts as well. This isn’t legal advice, and can’t substitute for a consultation with a lawyer about your specific circumstances. It focuses on U.S. law — the law may impose different requirements elsewhere. Still, we hope it helps you get started with confidence.

Why should I care? Copyright’s Sword of Damocles

In some circumstances, the operator of a platform that handles user content can be legally responsible for content that infringes copyright. That can happen when the platform operator is directly involved in copying or distributing the copyrighted material, when they promote or knowingly assist the infringement, or when they benefit financially from infringement while being in a position to supervise it. But these judge-made rules are often difficult and uncertain to apply in practice — and the penalties for being found on the wrong side of the law can be severe. Copyright’s “statutory damages” regime allows for massive, unpredictable financial liability. That’s why it’s important to limit your risk.

For Server Operators: Limiting Risk with the DMCA Safe Harbors

If you run a social network server, the safe harbor provisions of the Digital Millennium Copyright Act (DMCA) are an important way to limit your liability risk. The DMCA shields server operators from nearly all forms of copyright liability that can result from “storage at the direction of a user” — in other words, hosting user-uploaded content. But to qualify for this protection, there are steps a server operator has to take.

1. Designate A Contact To Receive Copyright Infringement Notices

First, you’ll need to provide contact information for someone who can receive infringement notices (a “designated agent”). That information needs to be posted in at least two places: on your server in a place visible to users (such as a “DMCA” page or post, or as part of your Terms of Service), and in the U.S. Copyright Office’s “Designated Agent Directory.” To post that information to the directory, you have to create an account at https://www.copyright.gov/dmca-directory/ and pay a small fee. The directory listings expire after three years, and once expired, your safe harbor protection goes away, so it’s important to keep that listing current.

2. Respond Promptly to Notices and Counter-notices

When you receive infringement notices, it’s important to respond to them promptly. Notices are supposed to identify the copyright holder, the copyrighted work they claim was infringed, and the post they claim is infringing. By deleting or disabling access to the posted material, you protect yourself from liability with respect to that material.

The theory behind Section 512 is that hosts don’t have to be in a position of deciding whether a post infringes someone’s copyright — it’s up to the poster, the rights holder, and potentially a court to decide that. A host who takes down posts whenever they receive an infringement notice is well-protected. But it’s equally important to recognize that hosts aren’t required to take down content in response to every notice. Infringement notices are frequently wrong, misguided, or abusive, or simply incomplete. Hosts who want to stand up for their users’ speech can choose to disregard infringement notices that seem suspect. While this risks losing the automatic protection of the safe harbor in each instance, it can still be done safely with careful preparation, ideally using a plan crafted with help from a lawyer. Bear in mind that people sending false notices, including by failing to consider whether a post is a fair use before asking a host to take it down, can be liable for damages under the DMCA.

The DMCA also allows the person who posted the material to send a “counter-notification” asserting that they really did have the right to post and that there’s no copyright infringement. Responding to counter-notifications is a good way for a host to demonstrate that they look out for their users. When a host receives a counter-notification, they should forward it on to the person who sent the original takedown notice and let them know that the post will be restored in 10 business days. Then, after that waiting period has elapsed, the host can restore the posted material. Just like with infringement notices, a host isn’t required to honor a counter-notification that appears to be fraudulent, but there’s no penalty for honoring it anyway.

3. Have A Repeat Infringer Policy

The next requirement is to have a policy of terminating the accounts of “subscribers and account holders” who are “repeat infringers” in “appropriate circumstances,” and to carry out that policy. Yes, that’s a vague requirement. It doesn’t require a “three strikes” policy or any other sports analogy. It just needs to be reasonable. Be sure your policy is spelled out in your website terms or “DMCA” page.

4. Don’t Ignore Known Infringement

Hosts need to take down user posts whenever the host actually knows that the post is infringing. In other words, a host isn’t protected if they ignore takedown notices based on technicalities in the notices, or if they learn about the infringement some other way. But hosts don’t need to actively look for infringement on their servers — only to act when someone notifies them.

5. Don’t Encourage Infringement

Finally, make sure that nothing you post or advertise actively encourages copyright infringement. For example, don’t post examples of users uploading copyrighted music or video without permission, or insinuate that your server is a good place for infringing content.

There are some other technicalities in the DMCA that can affect the safe harbor, which is why it’s always a good idea to consult with a lawyer. But following these steps will help protect you when you run a social media server — or any other kind of user-uploaded content platform.

For years, EFF has pushed technology companies to make real human rights commitments—and to live up to them. In response to growing evidence that Palantir’s tools help power abusive immigration enforcement by ICE, we sent the company a detailed letter asking how the promises in its own human rights framework extends to that work.

This post explains what we asked, how Palantir responded, and why we believe those responses fall short. EFF is not alone in raising alarms about Palantir; immigrants' rights groups, human rights organizations, journalists, and former employees have raised similar concerns based on reports of the company's role in abusive immigration enforcement. We focus here on Palantir’s own human rights promises.

At the outset, we appreciate that Palantir was willing to engage respectfully, and we recognize that confidentiality and security obligations can limit what it can say. Nonetheless, measured against Palantir's own human rights commitments, its decision to keep powering ICE with tools used in dragnet raids and discriminatory detentions is indefensible. A good-faith application of those commitments should lead Palantir to end its contract with ICE, and refuse new, or end current, contracts with any other agency whose work predictably violates those commitments.

Palantir’s Public Promises

Palantir has long said it performs comprehensive human rights analysis on its work. It has also worked with ICE for years, apparently in a more limited capacity than today. It has publicly embraced the UN Guiding Principles on Business and Human Rights, the Universal Declaration of Human Rights, and the OECD Guidelines for Multinational Enterprises. Additionally, in its response to EFF, Palantir says its legal responsibilities are only “the floor” for broader risk assessments.

That was the point of our letter. We asked what human rights due diligence Palantir conducted when it first contracted with ICE and DHS; whether it performed the “proactive risk scoping” it advertises, how it reviews work over time, what it has done in response to reports of misuse, and whether it has used “every means at [its] disposal”—including contract provisions, third‑party oversight, and termination—to prevent or mitigate harms.

For the most part, Palantir did not answer our accountability questions. It did correct one point: Palantir says it does not currently work with CBP, and available evidence supports that, though it also made clear it could work with CBP in the future.

Palantir also raised a red herring it often deploys in response to criticism. It denied building a 'mega' or 'master' database for ICE and denied creating a database of protesters, which some ICE agents have claimed to have been built. We call it a red herring because those denials sidestep the central issues: what capabilities Palantir's tools actually provide to ICE.

To be clear, EFF has never claimed that Palantir is building a single centralized database. Our concern is grounded in how Palantir’s tools allow ICE to query and analyze data from multiple databases through a unified interface—which from an agent’s perspective can be a distinction without a difference.

In the sections that follow, we compare Palantir’s account of its work for ICE with evidence about how its tools seem to be used, and explain why legality, internal process, and sustained “engagement with the institutions whose vital tasks exist in tension with certain human rights” are no substitute for real human rights due diligence—because respect for human rights must be measured by outcomes, not just process.

Palantir’s ICE Work Undermines Its Own Standards

Palantir says ICE uses its ELITE tool for “prioritized enforcement”: to surface likely addresses of specific people, such as individuals with final orders of removal or high‑severity criminal charges. But according to sworn testimony in Oregon, ICE agents use ELITE to determine where to conduct deportation sweeps, and the system “pulled from all kinds of sources” to identify locations for raids aimed at mass detentions, including information from the Department of Health and Human Services such as Medicaid data. A leaked ELITE user guide for 'Special Operations' also instructs operators to disable filters to "display all targets within a Special Operations dataset." Those details directly conflict with Palantir’s narrow description of ELITE’s role.

Additionally, Palantir's response leans on legal authority and the Privacy Act. But it does not identify any specific lawful basis for using Medicaid data in this way or explain how its software enables that access. Even if a legal theory exists, turning sensitive medical information into fuel for dragnet sweeps is hard to reconcile with its commitments to privacy, equity, and the rights of impacted communities. Its own human rights framework requires grappling with foreseeable harms its products may enable, not just invoking possible legal authorization.   

Reporting shows that many people detained by ICE had no criminal record, much less a serious one, and in many cases no final order of removal. An overwhelming percentage of those detained were, or appeared to be, from Central and South America, and nearly one in five ICE arrests were street arrests of a Latine person with neither a criminal history nor a removal order.

These facts raise obvious questions about discriminatory impact, racial profiling, and whether Palantir's tools are facilitating detention practices far broader than the company claims. Palantir's response does not meaningfully engage those questions, despite the company's commitments to non-discrimination and due process.

EFF’s letter asked Palantir to explain how it is honoring its commitments to civil liberties in light of reports linking Palantir-owned systems to facial recognition and other tools used to identify and target people engaged in observing and recording law enforcement, including in connection with the deaths of Renée Good and Alex Pretti. The letter also cites an incident in which an officer scanned protesters’ and observers’ faces and threatened to add their biometrics to a “nice little database.” Palantir’s response denies involvement in any such database.

A narrow denial about a single database does not answer the broader question: if ICE, its customer, claims it has this capability, what has Palantir done to ensure its tools are not used to chill protected speech, retaliate against observers, or facilitate targeting of people engaged in First Amendment‑protected activity? For a company that claims to value democracy and civil liberties, this is not a marginal issue; it goes to the heart of its human rights commitments.

Legality, Process, and Engagement with ICE Are Not Human Rights Standards

As mentioned above, Palantir leans heavily on legal compliance. It says government data sharing is “subject to, and governed by, data sharing agreements and government oversight” and that any sharing it facilitates is done according to “legal and technical requirements, including those of the Privacy Act of 1974.” It describes its role in ELITE as “data integration,” enabling ICE “to incorporate data sources to which it has access,” including data shared under inter‑agency agreements.

EFF is very familiar with the Privacy Act—we are suing the Office of Personnel Management over it currently. But Palantir’s response does not clarify how ICE legally has access to this information, how Palantir ensures that it follows those legal processes, or how Palantir’s software may have enabled access in the first place. More critically, that is still a legal answer to a human rights question, and legal compliance alone is insufficient as a human rights standard.

Human rights due diligence requires assessing foreseeable harms, responding to credible evidence of abuse, and changing course when the facts demand it—something Palantir, on paper, recognizes. That’s why it stresses that its legal responsibilities are only “the floor for [its] broader risk assessments,” pointing to the way it built toward GDPR‑style data protection principles and incorporated international humanitarian law principles before those requirements were formalized. If those commitments mean anything, Palantir has to explain how specific practices—like enabling ICE to use Medicaid data in dragnet raids—square with that broader standard.

Palantir also leans heavily on process. It points to a “layered approach” to risk, frameworks that purportedly examine multiple dimensions of privacy and equity, and “indelible” audit logs that track how its tools are used. Audit logs are not sufficient for protecting human rights. There is a long history of authoritarian regimes keeping extensive logs of their human rights abuses. Those structures can be useful for protecting human rights, but only if they are used to detect harm, trigger reassessment, and lead to changes in design, access, support, or contract enforcement when credible reports of abuse emerge.

That is why we pressed Palantir to spell out clearly what reports of misuse Palantir has received, what changes it made, and on what timeline. Again, instead of offering specific examples, Palantir points back to its internal framework and its willingness to “move towards the hardest problems” as evidence of effective efforts. But human rights are an outcome, not just a process.

Human rights due diligence is not a one-time approval at contract signing; under the UN Guiding Principles, it is supposed to be continuous, with new facts triggering reassessment. Complaints, media reports, leaks, litigation, and sworn testimony are exactly the kinds of events that should prompt review. If Palantir has an account for that work— how often it reviews ICE contracts, who conducts the reviews, what triggers them, and how findings reach the Board— it had every opportunity to describe it. Instead, it offered a generic assurance that it remains committed to human rights without engaging in the specifics. Confidentiality may sometimes limit disclosure, but it is no substitute for accountability.

What Needs to Happen Next 

Palantir wants credit for “mov[ing] towards the hardest problems” and engaging with institutions whose missions it says are “in tension with certain human rights” while having a human rights framework. But when the record includes violent raids, dragnet detentions, use of sensitive medical data, discriminatory targeting, retaliation against observers, and deaths tied to immigration enforcement operations, pointing to a values page is not enough; it has to reckon with the results.

Voluntary corporate human rights policies often function as weak accountability mechanisms: companies can tout principles, publish policies, and answer criticism with polished statements while changing very little on the ground. Palantir’s response fits that pattern all too well. EFF will continue to challenge its role in abusive immigration enforcement and demanding more accountability for technology vendors whose tools enable human rights violations. We are also happy to continue a dialogue with Palantir to that end. For now, this much is clear: Palantir needs to reconsider its contract with ICE and with all agencies whose work predictably violate human rights.