Mozilla, the non-profit behind the Firefox browser, is reinventing itself as a champion of open-source AI, in what its president Mark Surman says is an attempt to build a “Rebel Alliance” resisting the concentration of power in the tech industry.
On Tuesday, Mozilla released its first “state of open-source AI” report, which TIME viewed exclusively ahead of its publication.
The report argues that open-source AI has almost caught up to the “closed” frontier, and makes the case that more investment is needed at different parts of the ecosystem beyond the models themselves, where attention is usually focused.
As well as being a picture of the open-source ecosystem in 2026, the report is at least partially a piece of advocacy for Mozilla’s anti-concentration-of-power agenda, too, according to Raffi Krikorian, Mozilla’s chief technology officer. “I don’t want seven AGIs, one for every single one of the big companies. I want seven billion AGIs,” Krikorian says, referring to artificial general intelligence. “I want us to each have an AGI that’s truly on our side, not on someone else’s side.”
The report argues that the risk of power concentration in the AI industry doesn’t just reside in the models themselves, which are becoming easily interchangeable. Instead, the report argues, the open-source community must invest resources into other “layers” where power might otherwise accrue. Chief among them are so-called “harnesses,” which are programs that can amplify a model’s abilities, making it more useful and user-friendly. (Transferable memory, permissions, and pricing are all other areas that could do with more investment, Mozilla’s report argues.)
But some of the report’s assertions are questionable. A key statistic claims that open-source capabilities are just “3.3%” behind those of closed models—a number that the report later acknowledges conceals a “jagged” frontier, where closed models far outstrip open ones in important domains. The report also claims that open models are “at parity” with closed models on coding ability, albeit behind on reasoning—even though improved reasoning capabilities generally translate into stronger coding capabilities.. And many sections of the report contain telltale signs of AI writing, including Krikorian’s foreword, which the AI detection tool Pangram flagged as “100%” AI-generated. (In an interview, Krikorian said he had written the foreword himself, though acknowledged an editor may have used AI to edit it.)
The report barely acknowledges the risks of open-source AI. Because open-source models are made available to download, it is difficult to suppress them if they turn out to have dangerous capabilities. That applies to today’s risks, like deepfakes, and tomorrow’s, like the ability of forthcoming models to find cybersecurity flaws in the world’s software. When asked about the risks, Krikorian argues that progress in AI is essentially inevitable, and that protections must sit at the societal level. “If we just embrace the fact that these open models are racing forward,” he says, “[…] then we can have a real conversation on how we want to not necessarily contain, but use these systems [and] deploy these systems.”
The report contains clues about how Mozilla might be planning to sustain itself if the revenue streams from Firefox dry up. “You shouldn’t be surprised if in a few months you see that we’ve released our own harness,” Krikorian tells TIME. Surman, Mozilla’s president, says: “We’re really committed to this idea of the Rebel Alliance. We’re really committed to pulling together all of the different players who share that vision.”
Asked who the evil empire is in his Star Wars analogy, Surman replied: “I think the Empire is the idea of centralized, winner-takes-all tech. You have a number of companies who, that’s the game they’re playing.”