The post Era raises $11M to build a software platform for AI gadgets appeared first on Mozilla Ventures.

During our visit to the Barcelona Supercomputing Center last year we learnt about the innovative work that the Galtea team was doing in building quality assurance for AI systems. We returned to Barcelona for Mozfest 2025 and had the opportunity to spend time with the founders, Jorge Palomar and Dr Baybars Kulebi. We were struck by their conviction that without sufficient, affordable testing data, developers have no reliable way to know how their agents will perform in the real world. They had developed this insight as researchers at the Barcelona Supercomputing Center (BSC) where they saw firsthand the challenges of validating complex AI systems. 

Jorge and Baybars make a formidable team. Jorge is an AI Data Engineer turned Data Strategy Lead at the Barcelona Computing Center. He previously was a Business Intelligence Engineer at Amazon. Baybars is a physicist turned developer. He was previously the Engineering Manager at the Barcelona Computing Center. Together they spun Galtea out of the BSC and are now using high quality synthetic data to evaluate AI systems across three critical dimensions: quality, security, and real world user scenarios. Their platform simulates edge cases, stress tests agentic workflows, and generates use case specific evaluation datasets, giving enterprises measurable confidence before deployment. Early customers include Telefonica and leading Spanish and retail commercial bank, ABANCA, both of which use Galtea to dynamically test scenarios and tailored metrics for their AI agents, enabling them to make more informed decisions. 

Rather than treating evaluation as a secondary feature within observability, Galtea positions it as foundational. Without strong evaluation data, enterprises are guessing whether their AI is behaving correctly. As AI systems become more autonomous and agentic, that guesswork becomes riskier.At Mozilla Ventures, we invest in the infrastructure required for trustworthy AI. Galtea’s data first approach helps enterprises move from experimentation to production with greater safety and confidence, strengthening the foundations for AI that works, which is why we joined 42CAP in investing in the Galtea Seed round.

The post Building the Evaluation Layer for AI Agents appeared first on Mozilla Ventures.

The summit was the fourth in a series of global AI convenings (following Bletchley Park, Seoul, and Paris) and the first held in the Global South. But the real story for us wasn’t the declarations or the $200 billion in pledges. It was watching founders we’ve backed step onto the world stage and do the work.

The Portfolio in Action

Credo AI

Navrina Singh had a standout week. She spoke on “Scaling Trusted AI: Global Practices, Local Impact,” making the case that the challenge has moved from building AI to verifying and governing it.

Credo announced a strategic partnership with G42, the Abu Dhabi-based AI group, to operationalize responsible AI governance across the Global South and Middle East. Separately, the team launched a Global AI Governance Insights Hub, an interactive platform for tracking AI policy, risk controls, and regulatory mapping, and co-published a Responsible AI Compendium with G42.

In her reflections from the summit, Navrina distilled six takeaways that I think capture the state of play perfectly. Two stood out to me. First: trust is both the bottleneck and the unlock. Organizations will only scale AI when they can demonstrate measurable risk management and lifecycle governance. Second: as AI becomes agentic, static oversight and manual checklists can’t keep up. Continuous monitoring, real-time policy enforcement, and AI governing AI will define the next era. That’s exactly the infrastructure Credo is building.

Adaption Labs

Sara Hooker was on panels throughout the week on AI safety, evaluation, and cross-cultural robustness, including a Day 5 session alongside speakers like Ananya Birla and Aadit Palicha. She shared why the future of AI must move beyond static systems toward adaptation across real-world contexts. She also sat for a great interview on NDTV that captured both the big questions and the fun of the week.

Adaption also hosted their own social gathering in Delhi, bringing together researchers, policymakers, founders, and builders. As the team reflected afterwards: a clear takeaway was that policy conversations are accelerating: India is engaging seriously with how to shape AI’s future, not just scale it.

Sara’s work on ensuring AI systems work reliably across languages and cultural contexts was perfectly at home at a summit that put Global South priorities center stage.

Fiddler AI

Krishna Gade attended with his team, hosting an intimate community event and joining the Leaders’ Plenary. At one of Mozilla’s events, Krishna spoke about AI trust and transparency, making the case that observability must precede autonomy, that agents need a control plane, and that compounding workflows raise the stakes for governance.

In his summit reflections, Krishna noted what struck him most: India isn’t asking “how do we win the AI race?” but rather how to make AI accessible at population scale, avoid concentration risk, and ensure AI empowers small businesses and citizens rather than just incumbents. Having already built population-scale digital public infrastructure with Aadhaar and UPI, India is now thinking about a common AI substrate: interoperable, identity-aware, and secure. As Krishna put it: if India embeds trust into that AI substrate, it could shape a third path in global AI governance.

Lelapa AI

Vukosi Marivate spoke at multiple sessions throughout the week, bringing the perspective of building AI rooted in and designed for African languages and communities. His panel on “Collaborating to Scale AI Adoption in the Global South” was a natural fit for Lelapa’s mission, making the case that meaningful AI adoption requires locally rooted models, not just translated ones. He also joined an ORF/SAIIA dinner on inclusive AI governance, exploring how AI can reflect the diversity of human experience across languages, cultures, and identities.

Separately, Vukosi was named as one of 40 global experts appointed to the UN’s new Independent International Scientific Panel on AI. Selected from over 2,600 applicants across 140 countries, the panel is the first fully independent scientific body dedicated to assessing AI’s real-world impact across economies and societies. It’s a signal of how seriously the world is now taking the kind of inclusive, multilingual AI work that Lelapa represents.

Holistic AI

Raj Bharat Patel joined a panel co-hosted by techUK and BSI on AI assurance: “From Principles to Proof.” Chaired by Sue Daley, with fellow panelists Natasha Crampton and Tim McGarr, the session explored how assurance and governance can be drivers of deployment rather than barriers to it, and how UK-India collaboration can move the field forward at pace. In his reflections, Raj shared how Holistic AI is leading on governance deployment and emerging best practice.

Investing in Trustworthy AI

That was the portfolio. Here’s the bigger picture.

Our colleague Linda Griffin captured the summit dynamics well in her policy blog readout: at Bletchley, open source AI was treated as a security risk; at Paris, the consensus began to shift; at Delhi, there was undeniable recognition that open source is central to the AI future. Mozilla President Mark Surman told Fortune that beyond the photo ops, there was real hunger from countries, companies, and communities to build AI that is open-source, sovereign, and culturally tailored. As Mark put it heading into the summit: open source AI is the path to both economic and digital sovereignty.

I joined a panel with AI Commons on “Capital Allocations as AI Governance”, making the case that where we deploy capital shapes the kind of AI ecosystem we get. If we invest exclusively in closed, proprietary systems, we get concentration and dependency. If we invest in open, trustworthy AI infrastructure, we get resilience, sovereignty, and systems people can actually inspect and govern.

Mozilla CTO Raffi Krikorian made the same argument in Rest of World during the summit, and it’s the thesis behind Mozilla’s commitment to open-source AI capability development.

That was the theory. Walking out of that session and watching our founders do the work all week, signing governance partnerships, building assurance frameworks, making the case for observability and adaptation in rooms that matter, that was the practice. The best argument for investing in trustworthy AI turned out to be the portfolio itself.

The post At the India AI Impact Summit 2026: Our Portfolio in Action appeared first on Mozilla Ventures.

Deploying AI in the enterprise has led to impressive demos but when real traffic moves through enterprise systems, things start to behave differently. A workflow that looked stable in staging slows under load, or a dependency behaves differently than expected.

Today, the hard part is no longer just finding the right model but keeping the system around it from breaking.

AI in production means coordinating training, evaluation, deployment, monitoring, retries, logging, and versioning across probabilistic components. It means workflows that branch at runtime and failures that cannot be replayed from the beginning without incurring costs. That gap between experiment and production is where reliability is tested. 

Union.ai is building infrastructure for that gap. Mozilla Ventures backed Union.ai because a durable, open development infrastructure is foundational to Trustworthy AI.

The production reality

Many orchestration tools were originally designed for predictable workflows. The same input produces the same output. AI systems behave differently. 

They generate outputs that are non-deterministic. Agentic systems branch based on intermediate decisions. They interact with external systems in ways that are hard to predict.  

Teams are discovering that experimentation is no longer the bottleneck. Deployment and operationalization is. A few examples include a model passing validation but failing in deployment, or a workflow collapsing when a library version changes. Too often, debugging requires stitching together evidence across environments. At some point, the constraint is not the model itself but the layer coordinating everything around it.

Union.ai builds resilience into AI workflows and agents by making failures recoverable, and inspectable under real conditions.

Why this matters

Reliability at this layer affects more than developer experience. Systems that cannot be automatically recovered, versioned, or monitored are difficult to audit. 

Trustworthy AI is as much an operational challenge as a policy one. Operational trust depends not only on tooling, but on whether that tooling remains open to inspection. 

Open source as ecosystem strength

Union.ai was founded in 2020 as the enterprise AI orchestration platform behind Flyte. Flyte has surpassed 80 million downloads and powers AI orchestration for 3,500+ companies. Union.ai’s open-source portfolio also includes Pandera, a data validation framework with more than 100 million downloads. 

Mozilla’s open-source AI strategy emphasizes agency and choice. By building an open source orchestration layer, teams have visibility into how systems behave and how decisions are executed. That visibility makes accountability practical and increases resilience. 

A team working at the right layer

Ketan Umare, Union.ai’s co-founder and CEO, first encountered the limitations of traditional orchestration tooling while leading machine-learning engineering work at Lyft. There, Flyte, the open-source AI orchestrator at the heart of Union.ai, was developed to help disparate teams coordinate complex ML workloads that conventional tools weren’t designed to handle.

That conviction grew out of hard production lessons, where orchestration determined whether experiments ever became real systems.

“I founded Union.ai with a singular mission: to make orchestration of machine learning easy and repeatable, so teams of any size can turn experimentation into reliable production systems,” said Ketan. “Recognizing the gap between iterative ML work and production-grade infrastructure, we built Flyte to bring engineering rigor to AI workflows.”

Over years of real-world use across industry, from ride-hailing to media and financial services, Flyte’s open-source community grew as engineers adopted it for production workflows. That adoption, paired with persistent infrastructure friction in AI pipelines, inspired Union.ai to formalize and extend the tooling into a commercial platform that retains open-source DNA while helping teams move from experimentation to durable, production-grade systems.

Investing in Union.ai  

For Mozilla Ventures, we believe that building Trustworthy AI needs to be built across the development and deployment stack. That’s why we were excited to partner with Ketan and the Union.ai team and invested in their recent Series A led by NEA.  

The funding supports the commercial launch of Union.ai 2.0 and continued investment in supporting open-source AI.

The post Building the Infrastructure AI Actually Needs appeared first on Mozilla Ventures.

“We need a mindset shift so that we’re thinking more about technology as tools, and not as our masters,” said Katy Knight of Siegel Family Endowment. “There’s a very strong imperative from Big Tech to make us believe that the tech itself is the end. The reality is that we have the agency to choose what we adopt and what we don’t adopt.”

ImpactAlpha’s Agents of Impact Call brought together Knight, Chris Jurgens of Omidyar Network, Paul Fehlinger of Project Liberty Institute and Mohamed Nanabhay of Mozilla Ventures to sketch an investment thesis around responsible and ethical AI and kick off ImpactAlpha’s Shaping the Algorithm beat, in partnership with Siegel Family Endowment. 

A growing number of impact investors are getting comfortable with investing in the infrastructure of AI — AI governance models, “orchestration” layers, and assurance tech — in addition to AI-enabled applications. “If we don’t have that core AI stack safe and secure, then all the things we want to do for AI for good on top of it aren’t going to work,” said Jurgens.

Omidyar is working with partners to build some of the frameworks investors need to assess AI –  what Jurgens described as akin to a Taskforce for Climate-related Financial Disclosure to identify, disclose and engage on material AI risks. It’s also convening other foundations, family offices, pension funds and other asset owners on the issue. “There’s a really strong cohort of investors now that are ready to lead on this,” Jurgens said.

Omidyar itself is prepping a vehicle, Project Alexandria, that will invest in AI assurance and safety tech, such as  algorithm auditing and deepfake and other forms of fraud detection. “We need to treat that as an impact vertical, just as we would look at applications in health or climate as impact,” he said.

Fehlinger of the Project Liberty Institute, a nonprofit focused on building a people-centered digital future, has also been urging investors to invest in AI infrastructure, as opposed to just AI-enabled applications. “There’s a lot of need for innovation at the infrastructure level, the protocol level of the AI economy,” he said. 

Techno-pragmatism

Nanabhay of the open source nonprofit Mozilla zoomed in from India, where the India AI Impact Summit was taking place — and clashing notions of the AI future were on display. 

There was the usual excitement and optimism around AI as a productivity booster that can usher in new breakthroughs, Nanabhay reported. But there was also “a palpable sense of fear in the air as people think about some of the risks, think about job losses that are coming.”

Mozilla, an early thinker on building trustworthy AI systems, stood up a $35 million venture capital fund three years ago to invest in startups developing safe and inclusive AI. It has made some 55  investments to date. 

A self described techno-pragmatist, Knight staked out a middle ground between the doomers and the zoomers. She stressed the need to build brides between the AI for Good crowd and the Big Tech world that is raising vast sums of money in their race to dominate AI. “It can’t be competition. We will lose,” she said. “The scale of capital is just not there to serve every public good need. So how are we going to be savvy about making things greater than the sum of their parts, making something out of the work that we’ve been invested in across the social impact spaces we have cared about for a long time with AI and for AI in this moment, and not reinvent them?” 

For impact investors who have pursued good jobs, shared prosperity and a healthy planet, “It’s time for us to take these threads and really weave them together and think more deeply about the moment of opportunity.” 

Lively chat

Participants from foundations, venture funds, family offices, nonprofits and startups shared their takes in an active webinar chat sidebar. 

Katie Hallaran of Accion Ventures said that her team works with “early-stage inclusive fintech companies, helping them integrate AI in responsible ways.” Sorenson Impact Foundation’s Ibrahim Rashid reported that he has “taught a monthly workshop for 9+ months teaching investors how to build AI Agents to help with sourcing, diligence, reporting, and fund operations.”

A major thread in the chat centered on capacity gaps between the tech and impact sectors. Steven Clift put it bluntly: “The gap between the tech and impact world that Chris mentions is huge. In short, the impact world does not have the talent in-house to shape the direction of AI.” Nish Acharya of Equal Innovation added her two cents: “We are scouting AI startups for foundations and global NGOs around the world. Lots going on, but not as much as we would hope in this area.”

Another point of frustration: funding constraints for public-interest AI. Arclet’s Adrienne Ammermandescribed “how challenging it is to get funding for this kind of work” in sustainable, AI-enabled public health communications. Susan Bratton said that “to build a safe, effective, explainable, traceable, auditable platform… takes time,” adding that “the 5-7 year time frame isn’t necessarily right for that. [We] need longer money.” Astrid Scholz argued that the space “would benefit from more revenue based finance” to keep companies’ missions intact.

Equity and inclusion emerged as another focal point. Tanuja Prasad raised concerns about systemic bias, warning that AI’s probabilistic logic is “reinforcing the middle,” meaning “innovation and creativity are always in the minority.” Brett Kettle, who works in ecological modeling, agreed, and added that “so much of what is important are critical edge cases.”

The post (Human) agents of impact are shaping the algorithm for ‘good AI’  appeared first on Mozilla Ventures.

Here’s how our founders are showing up:

Adaption Labs — advancing critical conversations around AI safety, evaluation, and cross-cultural robustness to ensure systems work responsibly across languages and contexts. Here is where you can find Adaption this week in Delhi. 

Credo AI — championing trusted AI and governance frameworks that move responsible AI from principle to practice inside enterprises and institutions. Join Credo this week on the ground, schedule here!

Fiddler AI — driving transparency and AI observability, helping organizations monitor, explain, and operationalize trustworthy AI at scale. Learn more about Fiddler’s intimate event here!

Lelapa AI — building AI rooted in and designed for African languages and communities, ensuring global AI development reflects global voices. Come listen to Lelapa founder, @Vukosi Marivate at these events!

Holistic AI — enabling organizations to assess, mitigate, and manage AI risk with rigorous governance and assurance tools. Come hear @Raj Bharat Patel along with techUK discuss live here in Delhi!

Lastly, learn how to invest in Trustworthy AI with our own, Mohamed Nanabhay, Managing Partner at Mozilla Ventures discussing Capital Allocations as AI Governance here along with AI Commons. 

The AI Action Summit is more than a convening — it’s a global moment to shape how AI is built, deployed, and governed. We’re energized to see our portfolio companies contributing real solutions to ensure AI works for everyone.

If you’re in Delhi, connect with the teams — and follow along as we continue investing in founders building a more trustworthy, equitable AI ecosystem.

#AIImpactSummit #ResponsibleAI #TrustedAI #AIGovernance #AIForAll #MozillaVentures

The post See Mozilla Ventures at the AI Action Summit in Delhi 🇮🇳 appeared first on Mozilla Ventures.

For the past decade, AI progress has followed a simple formula: bigger models, more data, more compute. It worked. But the returns are diminishing and the costs keep compounding. Power, capital, and control have concentrated in a handful of labs that can afford the next doubling.

Meanwhile, the rest of us have become elevated prompt engineers. We contort our requests to fit the model’s limitations. We rephrase often, and we memorize recipes – “think step by step,” “you are an expert in…” – hoping the system will finally do what we actually need.

This relationship is backwards. Tools should adapt to people. Not the other way around.

Adaption is building AI that works the way it should: systems that evolve with the world, learn from context, and meet people where they are. Mozilla Ventures backed Adaption because this approach reflects a shared belief in open, community-driven technology. Learn more about Adaption on their website here. 

Introducing Adaption

Adaption was founded on a simple premise: intelligence is not static, and AI should not be either. Instead of one-size-fits-all architectures optimized for averages, Adaption is building flexible systems that can be tuned to specific environments and requirements. Today’s AI remains expensive to customize and slow to change. Adaption’s work centers on lowering those barriers through efficiency-first design, modular composition, and data-efficient alignment.

The mission is to build systems that evolve across languages, cultures, and local realities. 

Why this moment

Sara Hooker has been thinking about the limits of scale for years. Her recent essay, “On The Slow Death of Scaling,” lays out the case: scaling laws – the idea that performance predictably improves with compute – have become the industry’s central article of faith. But the evidence is shakier than the consensus suggests.

Scaling laws hold reasonably well for predicting pre-training loss. But when you measure actual downstream performance – the tasks people care about – the relationship gets murky. Some capabilities scale predictably. Others plateau, or emerge unpredictably, or don’t arrive at all. As Sara puts it: “The acceptance that there are emergent properties which appear out of nowhere is another way of saying our scaling laws don’t actually equip us to know what is coming.”

The frontier labs that bet everything on compute are likely under-investing in other directions. And those other directions are where the interesting work is now: gradient-free optimization, test-time compute, synthetic data that lets you reshape the training distribution on the fly, interface-algorithm co-design. These techniques can deliver 5-20x performance gains with a fraction of the compute.

This is the moment Adaption is built for. Not the end of progress – but a shift in where progress comes from. 

Why this team

Mozilla Ventures doesn’t back teams for generic “technical credibility.” We back founders whose work and worldview align with ours.

Sara Hooker grew up across Africa – Mozambique, Eswatini, Kenya, Liberia. She went to a Portuguese-language middle school.  For her, the gaps in AI language coverage aren’t abstract.

At Cohere, she led the Aya project: 3,000 researchers across 119 countries, building AI that works in 101 languages – more than 50 of which had never been served by generative AI. Time Magazine named her one of the 100 most influential people in AI.

Sara has been publishing on efficient adaptation, data pruning, and the limits of scale for years – before it was fashionable to question the scaling consensus. Her research on model merging, multilingual optimization, and synthetic data generation shipped in production systems like Aya Expanse.

Sudip Roy spent seven years at Google building ML infrastructure – Pathways, TensorFlow Extended – that powers thousands of production models, including Gemini. At Cohere he led inference: making AI work reliably and efficiently at scale. This team knows what to build and how to ship it.

AI ownership should not be a privilege

True AI ownership today is out of reach for nearly everyone. Developers build on distant monolithic models controlled by a privileged few. Customization requires steep expertise, expensive compute, and massive datasets. Builders who care about privacy, sovereignty, or local needs face a constrained choice between opaque APIs or limited local alternatives.

This centralization is economic as much as technical. It locks innovation behind unpredictable pricing, shifting vendor rules, and platform dependency.

Adaption is designed as both a technical and economic response built across three layers: continuously refining data to meet user intent, building intelligence that adapts across industries and specializations, and redesigning the human-AI interaction layer. Together, these make advanced AI capabilities more reliable and deployable in practice.

A bet on builders, not gatekeepers

Adaption is bringing together talent obsessed with new learning paradigms. It invites builders who want to reimagine the building blocks of AI progress.

Mozilla’s developer network, open-source heritage, and community-building experience are strategic assets that accelerate this flywheel.

In practical terms, this means a future in which a developer at a regional hospital, a startup in an emerging market, or a nonprofit working in local languages can build and own specialized AI models without relying on centralized APIs. 

Everything intelligent adapts

The last decade was defined by scale. The next will be defined by adaptability.

Mozilla Ventures backed Adaption because we believe the future of AI won’t be built by whoever has the most compute. It will be built by teams that figure out how to make systems learn efficiently, evolve continuously, and work for the people the monolithic models leave behind.

This is early,but a future where AI adapts to more people – to their languages, their businesses, their needs – is worth betting on.

That’s why Mozilla Ventures is proud to support Adaption.

The post Everything Intelligent Adapts appeared first on Mozilla Ventures.

The post Fiddler Raises $30M Series C to Power the Control Plane for AI Agents appeared first on Mozilla Ventures.

The post Mozilla is building an AI ‘rebel alliance’ to take on industry heavyweights OpenAI, Anthropic appeared first on Mozilla Ventures.

The result is our 2025 Mozilla Ventures Portfolio Convening Report — a snapshot of how founders across our portfolio are tackling today’s hardest technology challenges. From trustworthy AI governance and privacy-preserving systems, to rebuilding healthier social platforms, to navigating the emerging risks of AI agents, the report reflects the breadth of work happening across the portfolio.

It also highlights something equally important: who is building the future. Founders shared how gender, geography, and lived experience shape not only what gets built, but how resilient and relevant that technology becomes.

While openness remains a core throughline, it is not the only one. Across the convening, founders emphasized trust, accountability, operational rigor, and inclusive design as essential ingredients for building responsible technology that can scale.

We invite you to explore the report, learn more about our portfolio companies, and dive into the ideas shaping the next chapter of AI, security, and better social systems.

Read the report here.

The post Inside the Mozilla Ventures Portfolio Convening: How Founders are Meeting the Moment appeared first on Mozilla Ventures.