HumanSignal acquires Erud AI to build the world’s frontier data lab

AI has outgrown the internet.

The web gave us the data to start the AI revolution, but not the data to take it further. What comes next isn’t about scraping what already exists. It’s about building a new kind of data infrastructure, one that can reliably generate datasets each model or mission demands.

That's why today we’re announcing the launch of , alongside an acquisition of Erud AI, a company specializing in creating and evaluating domain-specific data, the kind needed to train and test new AI systems.

For the past decade, the internet has been AI’s library: a vast archive of what humanity has written, shared, and recorded, and that library was mostly text. The next frontier is multimodal: vision, audio, sensors, interactions, and it demands a new way to produce data. We call this next stage a data lab – a living system for manufacturing high-signal, purpose-built data, refined through expert labeling and rigorous evaluation.

With Erud AI, we’re bringing this lab online and on the ground, through a network of real-world physical data labs where domain experts, annotators, and engineers work side by side to design, collect, and evaluate data tailored to each use case.

• For frontier AI researchers, this means access to novel datasets continuously generated to your unique research requirements and experimental insights, fueling faster experimentation, safer alignment, and deeper control over emergent intelligence.
• For applied AI teams within enterprises, it means an extension of your workforce, teams of experts and evaluators who help benchmark models, validate results, and turn proprietary knowledge into structured intelligence. You already have the context, we bring the feedback and human-in-the-loop infrastructure.

Together, we’re building the infrastructure modern AI depends on, where every piece of data is designed, not scraped.

And at the center of it all is Label Studio – open source at its core, and built for flexibility and multimodality from the start. Its version extends that foundation to the scale and governance modern AI teams need, powering the creation, curation, and evaluation of the one thing AI cannot generate on its own: human signal.