Roblox has confirmed that an AI research agent built by one of its researchers mistakenly downloaded 145.5GB of 3D assets from community websites, including material hosted on adult-content platforms. The incident came to light after Open3DLab, a community for 3D artists, reported that a crawler had accessed thousands of project pages and downloaded more than 1,000 files from its sites and related platforms.
Roblox says the agent failed to operate as intended, has been decommissioned, and did not use the collected data for any purpose, including AI model training. The company also said it is strengthening safeguards for how AI agents are developed and reviewed. The incident highlights the risks of automated data collection, particularly when AI research tools encounter creator-owned assets, licensing restrictions and content that was never intended for a particular application’s users.
Key Takeaways
- 145.5GB of data: Open3DLab says the crawler downloaded 1,027 files from 712 projects across three community websites.
- Adult content included: The downloads included files from SmutBase, an adult-content site, and adult-marked projects on SFMLab.
- Roblox confirms the incident: The company said a researcher-built AI agent mistakenly scraped the content and has since been decommissioned.
- No AI training, according to Roblox: The company says the collected data was not used for training or any other purpose.
- Creator licensing concerns: Many downloaded projects reportedly carried licenses restricting commercial use, raising questions about permission and compliance.
- Broader AI safety issue: The incident demonstrates why autonomous data-collection agents need clear boundaries, monitoring and safeguards.
What Happened in the Roblox Scraping Incident?
Open3DLab reported that it detected unusually heavy automated traffic on October 18, 2026. The crawler identified itself with the user-agent string trellis2-data-sourcing/1.0, describing its activity as 3D foundation-model research and including a Roblox-associated email address.
According to the site’s technical account, the crawler accessed 18,926 project pages through 26,429 requests over approximately 34 hours. It subsequently downloaded complete copies of 1,027 files from 712 projects, totalling 145.5GB.
The affected platforms included:
- Open3DLab: A community hosting 3D models and related assets.
- SFMLab: A repository of models and resources used by digital artists and creators, including adult-marked projects.
- SmutBase: A site focused on adult-oriented 3D content.
Open3DLab said the crawler retrieved the sites’ robots.txt instructions but did not follow them. The operators also alleged that it ignored their terms of service and individual creators’ licensing restrictions.
Roblox subsequently confirmed that a researcher had built an agent to collect publicly available 3D assets. The company described the scraping as a mistake and said the agent had been decommissioned.
How Much Content Was Downloaded?
The total reported download was 145.5GB, but the material was not limited to general-purpose game assets.
| Reported activity | Volume |
|---|---|
| Project pages accessed | 18,926 |
| Requests made | 26,429 |
| Complete files downloaded | 1,027 |
| Projects represented in downloaded files | 712 |
| Total downloaded data | 145.5GB |
| Files from SmutBase | 364 |
| SmutBase data volume | 54.8GB |
| Adult-marked SFMLab data volume | 62.6GB |
Source: Open3DLab’s account, as reported by Tom’s Hardware. Figures are based on the site’s reported logs and have not been independently audited.
The reported volumes indicate that a substantial quantity of adult material was included in the collection. Open3DLab said it had no evidence that Roblox deliberately targeted adult content. Its concern was that the crawler appeared to collect assets without distinguishing between different content categories.
This distinction matters because the incident does not establish that Roblox intentionally sought explicit models, incorporated them into a product or exposed its users to them.
Roblox’s Response: The Data Was Not Used for AI Training
On October 28, Roblox confirmed that a researcher had built the agent to collect publicly available 3D assets and that it had mistakenly scraped content from Open3DLab.
The company’s statement said the agent had been decommissioned and the data had not been used for any purpose, including training AI models. Roblox also said it was strengthening safeguards governing how agents are built and reviewed.
That response addresses one of the central concerns: whether the downloaded files became part of a training dataset or influenced a deployed AI model. Roblox says they did not.
However, the distinction between downloading material and using it is important. The collection itself still raises questions about how the agent selected targets, how it handled restrictions and whether the safeguards were sufficient to prevent unwanted data acquisition.
Roblox’s public statement does not, by itself, resolve every question about the crawler’s configuration or its compliance with the individual licenses attached to downloaded files.
Why Were Creators Concerned About Licensing?
The controversy is not simply about the quantity of data collected. It is also about who created the assets and what permissions they granted.
Open3DLab reported that 423 of the 712 downloaded projects carried non-commercial licenses. Such licenses generally restrict commercial use, although the exact permissions depend on the terms attached to each asset.
A file being publicly accessible does not automatically mean it can be reused for every purpose. A creator may allow people to download an asset for personal projects while prohibiting commercial redistribution or other forms of use.
AI training adds another layer of complexity. Whether a particular use is permitted depends on the license, applicable law and the facts of the case. The scraping report alone does not establish that every download constituted copyright infringement.
Nevertheless, creators have a legitimate interest in understanding how their work is collected and whether automated systems respect the conditions attached to it.
For small communities, large-scale scraping can also create operational problems. Open3DLab said the crawler’s activity placed pressure on its download infrastructure and affected normal users’ ability to access files.
Why Was Roblox Collecting 3D Models?
Open3DLab suspected the activity was connected to Roblox’s research into generative 3D AI, particularly its Cube system.
Roblox introduced Cube as a foundation for generating 3D content, with applications involving digital objects, characters and interactive experiences. The company has also explored systems that can generate objects with motion and functionality from natural-language prompts.
The crawler’s identifying string referred to 3D foundation-model research, and Open3DLab said the email address in that string was associated with a Roblox researcher involved in the company’s 3D AI research.
These clues help explain why the site’s operators suspected an AI data-collection operation. Roblox’s later confirmation that a researcher built the agent supports the connection to the company.
However, the precise research purpose and intended dataset composition should not be inferred beyond the available evidence. Roblox has said the collected data was not used for any purpose, including model training.
What the Incident Reveals About AI Agents
The case illustrates a significant challenge as companies deploy AI agents to carry out tasks autonomously.
Unlike a simple script that downloads a fixed list of files, an agent may be designed to navigate websites, interpret information and continue working across many steps. If its instructions or safeguards are inadequate, it can collect material outside the intended scope before a person notices.
Three lessons stand out.
Public access is not unlimited permission. An agent must account for website rules, licensing restrictions and other relevant limitations rather than assuming that accessible files can be collected and reused without conditions.
Autonomy requires oversight. Developers should establish limits on what an agent can access, how much data it can retrieve and when its activity must be reviewed or stopped.
Content filtering matters. Data-collection systems need safeguards for sensitive or restricted material. The incident demonstrates why broad collection without adequate screening can produce results that were not intended by the people building the system.
These lessons apply beyond gaming. Companies developing AI models for images, video, audio, software and 3D environments face similar challenges when collecting data from public websites.
The Bigger Picture
The Roblox incident highlights the tension between rapid AI development and the rights of the communities whose work may become part of the wider digital ecosystem. Three-dimensional assets can represent substantial creative effort, and the terms under which creators share them are an important part of that relationship.
Roblox’s confirmation and decision to decommission the agent provide a clearer account of what happened than the initial allegations alone. Its statement that the files were not used for AI training is also an important qualification. Yet the event raises questions about how companies test autonomous agents, enforce website restrictions and prevent broad collection from crossing intended boundaries.
As AI agents become more capable, responsible deployment will require more than technically successful automation. Organizations will need to demonstrate that their agents respect access limits, creator permissions and content restrictions, with monitoring and review systems that can detect mistakes before they become large-scale incidents.
FAQs
1. Did Roblox really scrape 145.5GB of 3D models?
Roblox confirmed that a researcher-built AI agent mistakenly scraped content from Open3DLab. The site’s operators reported 145.5GB of downloads comprising 1,027 files from 712 projects.
2. Did the downloaded files include adult content?
Yes. Open3DLab reported downloads from SmutBase, which focuses on adult content, and adult-marked projects on SFMLab. The company said it had no evidence that Roblox deliberately targeted this material.
3. Did Roblox use the scraped models to train its AI?
Roblox says the agent was decommissioned and the data was not used for any purpose, including training AI models.
4. Was the scraping illegal?
The available reports do not establish a definitive legal finding. Open3DLab alleged that the crawler violated its website rules and ignored creator licenses. Whether particular downloads breached copyright or other laws depends on the relevant licenses, circumstances and applicable law.
5. What is Roblox Cube?
Cube is Roblox’s generative AI system for creating 3D content. It is part of the company’s broader effort to let creators build digital objects and interactive experiences using AI-assisted tools.
Looking Ahead
The next important question is how Roblox will strengthen its agent-development safeguards and prevent similar collection mistakes. Better target validation, license-aware filtering, explicit restrictions on sensitive content and human review of large-scale collection tasks could help reduce the risk of another incident.
For creators and AI developers, the case reinforces the need to distinguish between content that is publicly accessible and content that is authorized for a particular use. As automated data collection becomes more common, transparent practices and enforceable controls will be essential to maintaining trust between AI companies and the creative communities whose work they encounter.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
