Meta’s smart glasses just keep weirding people out. Renamed “pervert glasses” by some critics, the Ray-Ban smart glasses, made in partnership with EssilorLuxottica, have given rise to “creepy” behavior as users photograph and record people without their permission, then share those images and videos on social media and across the broader internet. The backlash points to a rapidly growing set of privacy, security, and ethical concerns.
At the same time, Meta is using the content generated across its ecosystem to support Mark Zuckerberg’s vision of a pervasive, AI-driven future. Zuckerberg’s 2026 manifesto describes a world where personal AI agents will do your bidding. But building those systems requires more than conventional AI models. It also requires enormous amounts of data about human behavior, much of it generated and shared through Instagram, Facebook, and other Meta apps, or captured through hardware such as smart glasses and EMG Neural Band devices. Privacy laws place limits on what Meta can collect, and in some contexts users can opt out. But vast quantities of data remain available, raising concerns that even seemingly incidental information can be swept into this system.
The pieces of Meta’s vision have been assembled over more than a decade. In September 2013, Facebook announced its first AI research lab to explore deep learning, and that December announced that it had hired the deep learning expert Yann LeCun, professor at NYU’s Courant Institute of Mathematical Sciences, to lead the effort. Six years later, in 2019, Facebook announced plans to integrate the technical infrastructure underlying WhatsApp, Instagram, and Facebook Messenger.
By the time of the Facebook Connect 2020 keynote, the company appeared to be “unifying underlying technical infrastructure” in service of a much larger ambition, as expressed by Zuckerberg:
“The road to the ultra-low-friction contextualized AI interface is a long and challenging one. But between EMG, egocentric data, and contextualized AI, I have not the slightest doubt that something a lot like what I’ve described will be how we will all work, play, and connect for however long the second wave of human-oriented computing lasts, which I think will be a very long time.”
During that keynote, Zuckerberg also described Project Aria, then framed as an AR research project and now also used for AI-focused behavioral research, while hinting at commercial smart glasses by recounting a 2019 trip to EssilorLuxottica in Italy to work on their design. The following year, in 2021, Meta launched Ray-Ban Stories. One month later, it announced its metaverse push, centered on augmented reality and virtual reality.
By 2022, that strategy was already struggling, just as ChatGPT and other large language models were about to bring generative AI into the mainstream. This year, Meta shuttered Horizon Worlds. Meanwhile, the company increasingly redirected its ambitions toward AI. It introduced its LLaMA model family in 2023, launched the standalone Meta AI assistant in 2025, and continued looking for new ways to put the data generated across its ecosystem to work. This year, Meta released Muse, another step toward AI systems capable of drawing on data from across that ecosystem. In that sense, Meta’s vision has not so much abandoned the metaverse as relocated it. The interface is increasingly the world around us, rather than a virtual world we choose to enter.
Zuckerberg’s goal of “low-friction contextualized AI,” built around technologies including EMG and egocentric data, ultimately relies on our labor. We generate the content, whether through Meta hardware, through other devices and then Meta apps, or simply by appearing in the photos, videos, conversations, and interactions of other people. Even those who do not actively use Meta products can still become data points when others record, photograph, mention, or communicate with them through these systems.
We become “algorithm chow,” feeding the models intended to realize Zuckerberg’s vision. Meta benefits at multiple points in that cycle, while users effectively help finance it through the forfeiture of privacy: by buying Meta hardware, using Meta apps, using services that interact with Meta, or simply becoming nonconsensual content captured by someone else. In that last case, opting out may be impossible. Maintaining privacy in a world of ubiquitous, networked cameras does not appear particularly scalable. We are all becoming part of the data source.
Mark Zuckerberg recently purchased an Irish castle that was formerly a prison before being destroyed and rebuilt. It makes for an almost too-perfect panopticon metaphor. Castles have become fashionable safe havens for the ultrawealthy. Outside their walls, meanwhile, the global commons risks becoming increasingly captive to Meta’s digital infrastructure, with ordinary people surrendering ever more privacy and security as Zuckerberg acquires ever greater means to preserve his own.