Mozilla

The latest data release from Common Voice dropped, marking a new milestone: Pashto, a widely spoken language that historically has been underrepresented in speech and AI datasets, is now the largest language dataset in Common Voice. Some release highlights: 10 new datasets 7 added languages 604 new hours of speech 593,000+ new speech clips Explore the new datasets at Mozilla Data Collective ⤵️ https://lnkd.in/ey_Tmu3e

Mozilla is proud to support the launch of the Open Source AI Fellowship, announced by the Open Source Initiative (OSI) during UN Open Source Week. As AI continues to shape our digital future, open source approaches can help foster transparency, innovation, collaboration, and public accountability. The fellowship will bring together researchers, policymakers, industry leaders, and civil society organizations to advance understanding of Open Source AI and its role in the broader technology ecosystem. We congratulate Gabriel T. on being selected as the Founding Fellow and look forward to supporting evidence-based dialogue, research, and policy development that strengthens openness and public benefit in AI. Thank you to Open Source Initiative (OSI), Duke University, and the broader community working to ensure AI remains accessible, trustworthy, and aligned with the values of openness. https://lnkd.in/eYuwkkFE

If we don’t laugh at the current state of technology, we’ll probably cry. So, let’s decide to laugh together! We are thrilled to partner with the brilliant Baratunde Thurston and The Onion's America's Finest Agency to explore our complex, messy, and deeply existential relationship with modern machines. In a landscape dominated by tech hype, Baratunde brings a much-needed reality check (and a lot of wit) to the conversations that matter most. Fittingly, our very first piece tackles the absurd, unsettling reality of employee tracking and workplace surveillance. Just this week, Meta paused an internal program designed to track employees' mouse clicks and keystrokes for AI training since it led to the accidental exposure of raw, private data company-wide. Read the debut column at the link in the comments below ⬇️

Who gets to decide when AI gets switched off? After the Trump administration ordered Anthropic to cut off access to its advanced cybersecurity AI models, the company shut them down for everyone — including trusted partners already using them. Raffi Krikorian joins Marketplace to discuss the implications of an AI “kill switch,” national security concerns, and who should have the authority to pull the plug. Listen here: https://lnkd.in/ehsnDk27

As AI adoption accelerates, one challenge is becoming impossible to ignore: where the data comes from — and who benefits from it. Mozilla Data Collective is proposing a new model for the AI data economy, one built around consent, community ownership, transparency, and fair value exchange. Rather than treating data as a resource to extract, the organization empowers creators and communities to decide how their datasets are used, shared, and compensated. With hundreds of curated datasets spanning more than 300 languages, Mozilla Data Collective is working to ensure underrepresented cultures and communities have a voice in shaping the future of AI. The initiative raises an important question for the industry: Can trust, sovereignty, and participation become competitive advantages in AI development? https://lnkd.in/eEkiQCmT

We loved this interview with SiliconANGLE & theCUBE. AI has a data problem: there isn't enough, and what there is was often created through extraction, opacity, and uneven value exchange. Some of these tactics are under legal scrutiny. We need to unlock new pathways to data abundance through working with, not around, communities. Mozilla Data Collective is building a different path: one rooted in community ownership, consent, trust, and fair value exchange. A place where dataset creators and stewards can share data on their own terms, and where developers can discover high-quality, contextualized datasets with greater transparency. This is exactly why Mozilla Data Collective exists: to support a more multilingual, multicultural, multimodal, and community-led AI future, built with human agency at the centre. 👉 Read the full feature with our CEO and Founder, E.M. here: https://lnkd.in/eEkiQCmT #AI #DataGovernance #ResponsibleAI #InclusiveAI

Can advanced AI make our digital infrastructure safer? Raffi Krikorian argues that the conversation around the shutdown of Anthropic's Mythos and Fable models is focused on the wrong question. The bigger issue is the absence of clear, accountable processes for disabling frontier AI systems—and whether restricting access actually improves security. Drawing on Mozilla's own experience using frontier models to identify and fix software vulnerabilities, Raffi makes the case that AI can be a powerful tool for defenders, researchers, and open source communities. If comparable capabilities are already spreading globally, the goal should be governance, transparency, and oversight—not simply the ability to flip a switch. A thoughtful contribution to the growing debate on AI safety, security, and public accountability. Shoutout @Transformer for covering this. https://lnkd.in/emNNkXps

Last call for builders of open source AI! The Developers Wilding track is looking for talks on production-ready AI stacks, agent security, local AI hardware, and the standards shaping the next generation of the ecosystem. If you're building outside the walled gardens and have lessons, experiments, or hard-earned insights to share, submit your proposal tonight. The deadline is June 15 at 2:00 PM CEST. Submit here: https://lnkd.in/gR9JM7XC