Short answer
Workplace recordings often capture content owned by others: branded packaging, product labels, software screens, signage, drawings, posters, background music and radio. A site owner can usually license only its own recordings, not rights held by third parties in what appears in them. Buyers should ask what third-party content appears, how it is handled, and who warrants what.
This page is about diligence questions, not legal conclusions. This is not legal advice; ask qualified counsel about your jurisdiction and your use.
The buyer's problem
A supplier can grant what it holds. A warehouse operator holds rights in its footage as a recording, subject to its agreements with employees, customers and landlords. It does not necessarily hold rights in the logo on a customer's carton, the drawing taped to a workstation, the manual open on a tablet or the song on the shop radio. Those belong to others, and they appear in training data by accident rather than design.
Whether and how third-party content matters depends on the content, the use and the jurisdiction. The point for procurement is narrower: find it, classify it, and decide in the contract who bears the risk. For people in footage, see worker consent and identifiable people. For customer and business secrets, see confidential business information in footage.
What third-party content shows up
| Content | Where it appears | Typical modality |
|---|---|---|
| Brand names, logos, trade dress | Packaging, uniforms, trucks, equipment | Video, images |
| Product designs and labels | Items being handled | Video, images |
| Documents and drawings | Work instructions, posted specs, clipboards | Video, images |
| Software interfaces | Handheld scanners, monitors, tablets | Video, screen capture |
| Music and broadcast audio | Radios, speakers, phone calls | Audio |
| Spoken content | Customer or supplier calls overheard | Audio |
| Photos and art | Posters, calendars, wall displays | Video, images |
| Third-party machine and tool software | Controller displays | Video |
Audio deserves particular attention. Video can be delivered with audio removed; an audio stream may carry music, conversations and other recordings, and removing it is simpler than cleaning it. Ask whether the audio track is needed for your use at all.
What the U.S. Copyright Office report covers
The U.S. Copyright Office released a pre-publication version of "Copyright and Artificial Intelligence, Part 3: Generative AI Training" on 9 May 2025. It addresses the use of copyrighted works in developing generative AI systems, identifies points in development where copying may implicate rights (Section III), analyzes fair use (Section IV) and examines licensing options (Section V) (U.S. Copyright Office, 2025).
Three cautions for a robotics buyer. The report concerns generative AI training, and a robot-learning model may or may not fall inside that analysis. It is a U.S. agency report, not a statute or a court ruling, and it is a pre-publication version. And fair use is a case-by-case analysis; no page, including this one, can tell you a particular dataset is covered. Take the report as a map of questions, and take specific conclusions to counsel.
Open licenses on embedded content
Some background material may carry public licenses. Creative Commons describes BY-NC-ND as its most restrictive license (unadapted copies only, noncommercial, with attribution), explains NonCommercial as not primarily intended for or directed towards commercial advantage or monetary compensation, judged by the use rather than the user, and notes that licenses with NC or ND terms are considered non-free (Creative Commons). If a supplier says a poster, image or audio clip is "open licensed", ask which license, whether it permits commercial use and adaptation, and whether the supplier can show it. For datasets, a similar caution applies to public benchmarks; see public dataset commercial license checks.
Handling options and their trade-offs
| Option | What it does | Cost to training value |
|---|---|---|
| Exclude affected clips | Removes the content entirely | Reduces coverage; may bias toward bland scenes |
| Blur or mask regions | Hides logos or text | Can remove cues a perception model needs |
| Strip or replace audio | Removes music and speech | Loses audio if you need it |
| Seek a license or release | Gets permission | Time and cost; uncertain |
| Contractual allocation | Supplier warranty and indemnity, buyer risk acceptance | Does not change the underlying rights |
Blurring or masking changes what a model sees; it does not by itself settle whether the content could be used. Treat it as a mitigation to be discussed, not a resolution. Manufacturing and warehouse perception tasks often depend on reading labels and packaging, so masking may defeat the purpose; see product catalog data for robot perception.
Illustrative third-party content register
# Illustrative only - synthetic values; a register a supplier could complete
site: site_017
categories_present:
- brand_marks_on_goods: yes, customers client_A, client_B (synthetic)
- posted_drawings_and_work_instructions: yes, 3 stations
- software_screens: yes, handheld scanners
- background_music_or_radio: intermittent
- overheard_calls_or_conversations: yes
handling:
audio: removed
screens: masked where legible
drawings: stations excluded or masked
rights_position:
site_owner_grants: recordings and its own materials
third_party_content: not licensed; flagged by category per clip
customer_permissions: obtained for client_A, not obtained for client_B
Questions for the contract
Clause topics to raise with counsel (not a template):
- a warranty from the supplier about what it can license, with scope limits stated;
- a representation about whether customers' goods, labels or confidential items appear;
- notice and takedown mechanics, and what happens to trained models if an item must be removed; see model weights, derivatives and redistribution;
- indemnity scope and caps, and who controls defense;
- a list of excluded categories and a process to add exclusions;
- audit rights to inspect the register; see retention, deletion and audit rights.
Limitations
Registers depend on reviewers recognizing third-party content, and reviewers miss things: a small logo, a drawing seen briefly, music at low volume. Automated detection helps but is imperfect. The review burden grows with hours of footage, which affects cost. No procedure makes the risk zero; the aim is to find the high-visibility content, document the position and allocate residual risk deliberately. Related tension with permissions for people is discussed in licensing terms for robotics training data.
Questions to ask suppliers
- Which categories of third-party content appear in the data, and by what review method were they identified?
- Are customers' brands, goods or documents visible, and do any customer contracts restrict recording?
- Is the audio stream included, and what is on it?
- Which on-screen software or screens are legible, and how are they handled?
- Is any music or broadcast audio present, and has it been removed?
- What claims are made about open-licensed material, and what evidence supports them?
- What will the supplier warrant, and what exclusions apply?
- How will flagged items be removed on request, and in what time frame?
- Can the supplier provide a clip-level register of flagged content?
Discuss your requirements
If third-party content matters to your use, describe the modalities, exclusions and licensing requirements to SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. A request does not guarantee a match; SourceX follows up about potential sources. See the humanoid training data guide and the procurement hub.
Explore sourcing and licensing options with SourceX
Sources
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) — U.S. Copyright Office pre-publication Part 3 report on generative AI training: sections on copying, fair use and licensing.
- About CC Licenses — Creative Commons license descriptions: BY-NC-ND, NonCommercial and NoDerivatives terms.