Bluesky ATProto Trademark: A Practitioner's Guide for 2026

I got the email in March 2024. A client was building a social graph analyzer on the AT Protocol, and their legal team flagged a USPTO filing by Bluesky, PBLL...

bluesky atproto trademark practitioner's guide 2026
By Nishaant Dixit
Bluesky ATProto Trademark: A Practitioner's Guide for 2026

Bluesky ATProto Trademark: A Practitioner's Guide for 2026

Free Technical Audit

Expert Review

Get Started →
Bluesky ATProto Trademark: A Practitioner's Guide for 2026

I got the email in March 2024. A client was building a social graph analyzer on the AT Protocol, and their legal team flagged a USPTO filing by Bluesky, PBLLC. Application 97/619,381 for “AT PROTOCOL” – standard character mark, intended use in Class 9 (software) and Class 42 (cloud services). My first reaction: this is a branding problem, they'll just rename the protocol. Turns out I was wrong. It’s a design constraint, a governance trap, and a licensing minefield all rolled into one.

I’m Nishaant Dixit, founder of SIVARO. We’ve been building production AI systems on decentralized data infrastructure since 2018. The Bluesky ATProto trademark isn’t just a lawyer’s headache. It affects how you architect relay nodes, handle user DIDs, and even how you spin up GPU clusters for inference on federated data. By the end of this guide, you’ll understand exactly what the trademark covers, why it matters to anyone running a production service on ATProto, and how to navigate the 2026 landscape without getting a cease-and-desist.

The Day the Trademark Hit My Inbox

Back in 2023, Bluesky’s open protocol was the darling of the decentralized social crowd. Everyone was forking it. I was experimenting with a custom PDS (Personal Data Server) for a CI/CD pipeline that needed bot accounts. Then in November 2023, the filing became public. The community erupted. "How can you trademark a protocol?" "This is antithetical to decentralization." I remember a Hacker News thread with 900+ comments – mostly rage.

But here’s the part most people miss: the trademark isn’t on the protocol specification. It’s on the word mark “AT Protocol” for specific goods and services. The actual protocol – the data model, the Lexicon schemas, the transfer protocols – remains Apache 2.0 licensed. The trademark gives Bluesky the right to stop you from labeling your own incompatible implementation as “AT Protocol”.

That distinction is critical. If you build a compliant ATProto relay and call it “My Super AT Protocol Server”, you’re in trouble. If you call it “My Super Atmospheric Social Relay”, you’re fine. Most people think the trademark kills the protocol. It doesn’t. It just forces you to think about naming and branding differently.

What the Bluesky ATProto Trademark Actually Covers

Let’s get specific. The trademark application (serial 97/619,381, filed August 2023) lists:

  • Class 9: Downloadable software for building and operating decentralized social networks
  • Class 42: Platform as a service (PaaS) featuring computer software platforms for decentralized social networking

The mark is “AT PROTOCOL” in standard characters. That means no stylized font, no logo – just the words. As of mid-2026, the mark is registered (USPTO registration 7,315,496, issued June 2025). Bluesky has also filed international trademarks via the Madrid Protocol in the EU, UK, and Japan.

What does this cover? It does not cover:

  • The protocol implementation (source code is Apache 2.0)
  • The underlying data structures (ATProto uses CBOR and IPLD, both open)
  • The concept of a decentralized social protocol

What it does cover:

  • Using “AT Protocol” as a brand name for your software product
  • Marketing your platform as “AT Protocol compatible” in a way that implies endorsement
  • Registering domain names like atprotocol-saas.com for commercial services

Practically: if you build a relay, call it something else. I’ve seen people use “ATRelay”, “BlueRelay”, “ProtocolBridge”. No one’s been sued. But if you put “AT Protocol” in your app name, expect a letter.

Why Decentralized Protocols and Trademarks Clash

Here’s the friction. Protocols thrive on network effects. The value of ATProto increases as more people build interoperable services. Trademarks, by design, create monopoly over identifiers. These two things don’t naturally align.

I’ve spent years architecting data infrastructure. When we built GPU clusters for real-time inference on federated data, I learned that naming consistency matters. You need clear labels for relay nodes, DIDs, and service endpoints. A trademark that restricts those labels introduces overhead. Every developer now has to check: does my service name infringe?

The real pain point is not for hobbyists. It’s for enterprises. I’ve talked to three Fortune 500 companies in 2025 who were evaluating ATProto for internal communication tools. All of them asked: “If we build an enterprise PDS and call it ‘AT Enterprise Server’, do we need a license?” The answer is complicated. Bluesky has not released a formal trademark licensing policy as of July 2026. The community has been pushing for a “trademark guidelines” document similar to what Kubernetes or Apache Foundation provide. None exists yet.

How the Community Reacted – And What Changed

The backlash in late 2023 was intense. But something interesting happened. Instead of forking the protocol (which would have been technically trivial given the Apache license), the community started building around the trademark creatively.

In January 2024, the “AT Protocol” GitHub org became “ATProto” (no space, capital P). Developers started using “ATProto” universally. The official spec website now uses atproto.com (no “protocol”). Bluesky itself never enforced the trademark against non-commercial projects – I know of zero lawsuits. But the chilling effect was real. Several venture-backed startups shelved ATProto-based products because their counsel couldn’t get clear trademark assurances.

By 2025, Bluesky did something smart. They published a FAQ page (still live in 2026) stating that:

  • Non-commercial use of “AT Protocol” in descriptions is permitted
  • Commercial use requires a license from Bluesky (contact [email protected])
  • They will not assert against use that “clearly does not confuse consumers”

That last clause is vague. In practice, I’ve seen zero licenses issued publicly. The approach is: don’t ask, don’t tell.

Practical Implications for Developers Building on ATProto in 2026

Practical Implications for Developers Building on ATProto in 2026

If you’re reading this because you’re building a product on ATProto, here’s my advice from three years of doing it.

Naming Your Service

Do not include “AT Protocol” in your product name. Use “ATProto” or “BlueSky” or something clearly distinct. I’ve shipped a PDS-as-a-service called “RelayStack”. Zero issues. A friend called his analytics tool “AT Protocol Insights” and got a polite email from Bluesky’s lawyer within two weeks.

Repositories and Packages

Your GitHub repo title is fine as long as the repo name doesn’t imply official status. atprototools/client is riskier than blue-client. If you’re publishing an NPM package, avoid @atprotocol/ scope. Use @atproto/ (lowercase) – that scope is owned by Bluesky and they’ve allowed community packages.

Documentation and Marketing

You can say “compatible with the AT Protocol” in your documentation. That’s nominative fair use. I’d avoid putting it in your page title or header logo. One company I consulted for put “AT Protocol Compatible” in their hero banner – they got a take-down request within 48 hours.

License Your Own Trademark

If you build something popular, consider trademarking your own brand. The ATProto trademark only covers “AT Protocol”. So I registered “RelayStack” as a trademark in Class 42 in 2025. Gives me leverage if Bluesky ever tries to extend their claim.

The Technical Side: Byzantine Agreement, Blocklace, and Trademark

Now we get to the part that fascinates me as an engineer. The trademark issue interacts with ATProto’s underlying architecture in ways most legal teams don’t see.

ATProto’s data model uses what’s called a blocklace – a conflict-free replicated data type (CRDT) structure that allows concurrent edits. Understanding byzantine agreement round complexity with an adaptive adversary is critical when designing relay consensus mechanisms. But how does a trademark affect that?

Consider this: the AT Protocol spec defines a specific way to structure blocks using content-addressed hashes. If a fork emerged that changed the blocklace structure (say, to improve CRDT convergence under adversarial conditions), would it still be “AT Protocol”? The trademark gives Bluesky the power to say “no” – even if the fork is technically compatible at the data level. This is exactly what happened with the blocklace CRDT attack scenario I worked on in 2025. We found a new type of conflict resolution that was faster but used a different merge function. Calling it “AT Protocol” would have been trademark infringement, even though we used the same repo codebase. We named it “ConfluxRelay” instead. It works fine, but we lost discoverability.

The deeper point: trademarks freeze the semantic boundary of a protocol. ATProto’s design encourages experimentation – anyone can run a relay, modify the data model, deploy a custom PDS. But if your modification touches the branding, you either hide it or rename it. That creates a tax on innovation.

Lessons from Other Decentralized Projects

This isn’t new. Mastodon has “Mastodon” as a trademark owned by Mastodon gGmbH. They have clear guidelines: you can run a server, but you can’t call it “Mastodon Enterprise” without permission. Ethereum doesn’t trademark the word “Ethereum” – the Ethereum Foundation has a trademark guideline that allows forks to use the name only if compatible. Matrix.org uses “Matrix” as a trademark, but their trademark policy is liberal for non-commercial use.

ATProto sits somewhere in between – more restrictive than Matrix, less than Mastodon. The problem is ambiguity. Mastodon published their policy in 2017. ATProto still hasn’t. Three years after the filing, Bluesky owes the community a clear, machine-readable license.

If you’re building on top, assume the worst. Plan to brand everything yourself. That’s what we did at SIVARO. Our production AI systems ingest data from multiple ATProto relays – we never refer to them as “AT Protocol infrastructure” in any external communication. Our clients don’t care. They care about reliability and latency.

Building a Production AI System that Depends on ATProto – My Approach

At SIVARO, we run a cluster of 32 Nvidia A100 GPUs for real-time content moderation on social data. ATProto is our primary data source. Here’s how trademark considerations hit operational reality.

We spin up PDS instances for testing – not user-facing, just internal. Should those PDS names include “AT Protocol”? Our ops team wanted to call them at-proto-test-01. I said no. It’s not worth the risk, even for internal. We use svr-pds-01.

When we deployed a custom GPU cluster for training sentiment models on ATProto feeds, we needed to document the data pipeline. The marketing team wanted a case study titled “Processing AT Protocol Feeds with 200K events/sec”. I had to kill it. Changed to “Processing Decentralized Social Feeds”.

Is that paranoid? Maybe. But I’ve seen the letters. Our legal cost for trademark review on one product launch exceeded $12k. That money could have bought us another GPU node on Vast.ai. Don’t underestimate the hidden cost of ambiguity.

Code Example 1: Checking ATProto Trademark Status via USPTO API

`python
import requests

# USPTO trademark status API (public)
serial = "97619381"
url = f"https://tsdr.uspto.gov/tsdr/api/v1/trademark/{serial}"
resp = requests.get(url)
data = resp.json()
status = data['cases'][0]['caseFile']['trademarkCase']['caseData']
print(f"Mark: {status['mark']['markLiteral']}")
print(f"Status: {status['trademarkCaseStatus']['statusDescription']}")
# Output in 2026: AT PROTOCOL, Registered

Code Example 2: Checking a Service Name for Potential Infringement

`python
# Simple heuristic - not legal advice!
def check_infringement(name):
    dangerous_phrases = [
        "at protocol", "atproto", "bluesky protocol", "bsky protocol"
    ]
    lower = name.lower()
    for phrase in dangerous_phrases:
        if phrase in lower:
            return f"RISK: Contains '{phrase}'"
    return "Probably safe, but consult trademark attorney"

Code Example 3: ATProto Relay Configuration – Avoiding Trademark in Logos

`yaml
# relay config example - safe naming
relay:
  host: "relay.example.com"
  port: 443
  # never use 'atproto' in hostname
  handle: "example-relay"
  # lexicon schema from official repo (Apache 2.0) - fine to use
  lexicon: "https://github.com/bluesky-social/atproto/main/lexicons/app/bsky/feed/getFeed.json"

Code Example 4: Verifying DID Document Compliance

`json
{
  "@context": ["https://www.w3.org/ns/did/v1"],
  "id": "did:plc:abcd1234",
  "alsoKnownAs": ["at://example-relay.com"],
  // Good – no trademarked phrase in alsoKnownAs
  "service": [{
    "id": "#atproto_pds",
    "type": "AtprotoPersonalDataServer",
    "serviceEndpoint": "https://pds.example.com"
  }]
  // "type" uses "Atproto" not "AT Protocol" – safe per community guidelines
}

Code Example 5: License Compatibility Scanner for ATProto Dependencies

`bash
#!/bin/bash
# Check all repos in current directory for Apache 2.0 + custom trademark notice
for repo in */; do
  if [ -f "$repo/LICENSE" ]; then
    if grep -q "Apache-2.0" "$repo/LICENSE"; then
      echo "$repo – Apache 2.0 (safe)"
    else
      echo "$repo – custom license (review trademark clause)"
    fi
  fi
done

Conclusion: The Trademark Game is Not Zero-Sum

After three years, I’ve made peace with the Bluesky ATProto trademark. It’s not an attack on open source – it’s a defensive move by a startup that doesn’t want its brand diluted. The real problem isn’t the trademark itself; it’s the lack of a clear licensing policy. Compare to Kubernetes, where the CNCF publishes explicit trademark usage guidelines. Bluesky needs to do the same before they scale further.

For now, my advice is pragmatic: assume any use of “AT Protocol” as a brand identifier will be contested. Build your own identity. The protocol is open. The network is growing. I’ve seen ATProto handle 200K events per second through SIVARO’s pipelines. It works. Just don’t call it “AT Protocol” in public.

FAQ

FAQ

1. Can I use “AT Protocol” in my open-source project name?
Technically no, but enforcement is unlikely for non-commercial. Still risky. Use “ATProto” or a unique name.

2. Does the trademark prevent me from writing “compatible with the AT Protocol”?
Nominative fair use typically permits accurate descriptive statements. Avoid placing it in logos or product names.

3. Has Bluesky sued anyone over the trademark?
As of July 2026, no public lawsuits. Cease-and-desist letters have been sent to at least two commercial SaaS products (according to community reports).

4. Can I fork the AT Protocol source code and call it something else?
Absolutely. Apache 2.0 license grants full rights. Rename it, rebrand it. Just don’t call it “AT Protocol”.

5. What about using “AT Protocol” in academic papers or documentation?
Safe. Trademark law allows referential use. Cite it as “the AT Protocol” in text. Avoid putting it in your publication’s title if the paper itself is a commercial product.

6. Is there any public trademark license from Bluesky?
Not yet. They’ve promised guidelines for over a year. The community is frustrated. I’d expect something by end of 2026.

7. How does the trademark affect data infrastructure like GPU clusters?
Indirectly. If you label your cluster “AT Protocol GPU Cluster” you’re infringing. I suggest using generic descriptions like “decentralized social network feeder cluster”.

8. What’s the worst-case scenario for a developer using the trademark incorrectly?
A cease-and-desist, potential rebranding costs, legal fees. I’ve seen small projects forced to change domain names and all documentation. Expensive.


Nishaant Dixit — Founder of SIVARO. Building data infrastructure and production AI systems since 2018. Built systems processing 200K events/sec.

Free · No Commitment · 48-Hour Delivery

Get a free infrastructure audit

2-hour remote session. We audit your data infrastructure, identify what's costing you time and money, and deliver a written roadmap with specific, measurable targets. No pitch.

Book Your Free Audit
N
Nishaant Dixit
Founder & Lead Engineer at SIVARO

Building data-intensive systems since 2018. 200K events/sec pipelines, production RAG systems, Kubernetes infrastructure. LinkedIn →

Start a Project
Need help with your infrastructure?

From data platforms to AI systems — we build production-grade infrastructure that scales.

Explore Our Services