The pitch, when you strip away the technical jargon, is straightforward enough: How do you know your AI agent won't go rogue before you unleash it on customers?
For Galtea, a startup that emerged from Barcelona's prestigious Supercomputing Center, that question has become the foundation of a fledgling business—and the hook for a $3.2 million seed round announced late last month. Mozilla Ventures and Munich-based 42CAP co-led the March 25 financing, joined by Abac Nest Ventures, JME Ventures, and Masia, a collection of investors betting that the unglamorous work of testing AI systems will become essential infrastructure.
Perhaps more essential than the founders initially expected. Jorge Palomar Giner and Dr. Baybars Külebi, both alumni of the Barcelona Supercomputing Center where they conducted research, spun out the company in 2024 with backgrounds that blend academic rigor and practical engineering. Palomar previously led data strategy at BSC and spent time as a business intelligence engineer at Amazon; Külebi made the unusual leap from physicist to engineering manager at the research center, a trajectory that speaks to the interdisciplinary scramble happening across AI labs.
The Compliance Crunch
What Galtea sells is less exciting than the models themselves but arguably more critical: systematic evaluation infrastructure for generative AI agents before they hit production environments. The platform automatically generates test scenarios and evaluation data, focusing heavily on regulatory compliance—particularly Europe's AI Act—and risk assessment protocols that enterprises are now racing to implement.
It's a bet on anxiety, essentially. Companies deploying GenAI systems face a thicket of regulatory requirements and reputational risks, and few have internal mechanisms to rigorously test non-deterministic systems that behave differently each time they run.
Mozilla Ventures' announcement of the investment noted that Galtea has already landed early customers including Spanish telecom giant Telefónica and ABANCA, a regional bank. The Telefónica relationship runs deeper than a typical customer engagement—Wayra, the telecom's corporate venture arm, put money into Galtea during 2024, establishing a connection that likely smoothed the path to commercial deployment.
A Crowded Niche
Galtea enters a market that's becoming conspicuously less empty. Investor appetite for AI evaluation infrastructure has accelerated sharply over the past year or so. Fiddler AI pulled in $30 million in Series C funding this past January, with Mozilla Ventures participating there as well. The company positions itself as a "control plane for AI," offering evaluation and security capabilities. Earlier rounds painted a similar picture: Galileo raised $45 million in Series B in late 2024, while Arize AI closed a $70 million Series C in February of last year.
The timing isn't coincidental. Academic researchers have increasingly turned their attention to the technical challenges of testing AI agents. Multiple papers published on arXiv in recent weeks have examined agent regression testing, coverage metrics, and continuous integration workflows tailored to non-deterministic AI systems—problems that were largely theoretical a few years ago and are now painfully practical for engineering teams.
Opening the Gates
Alongside the funding announcement, Galtea launched a self-serve "Start for free" tier, a bid to get developers experimenting with the platform before enterprise sales teams come knocking. The company's marketing leans heavily on compliance reporting and audit-ready documentation, features that sound bureaucratic but matter enormously for companies navigating the evolving regulatory landscape through 2026 and beyond.
The Barcelona-based team holds ISO 27001 certification and frames its technology as evaluation infrastructure rooted in research from BSC's Language Technologies Unit, built in the MareNostrum 5 high-performance computing environment. That academic pedigree may prove valuable as enterprises look for credible partners—though credibility and a few million in seed funding don't guarantee survival in a market where deeper-pocketed competitors are already entrenched.
Still, the founders are making a calculated wager: that the gap between building AI agents and safely deploying them will only widen, and that someone needs to fill it. Whether Galtea can carve out defensible ground remains the open question.
