AI agents get smartglasses, sellers' tools — and a first data breach

Mark Zuckerberg stood on stage this week and told the world not to worry so much about AI doom, then proceeded to launch a wearable AI agent that lives on your face. There's something almost poetic about that juxtaposition, and I don't think it was accidental.

Meta's Connect conference gave us Muse, an AI agent baked into a new generation of smartglasses, plus a handheld device Zuckerberg is positioning as an early step toward what he calls "personal superintelligence." Strip away the branding and what you actually get is Meta betting that the next computing platform is worn, not carried, and that an always-listening, always-watching agent is the interface people will want. Maybe. But the privacy backlash was immediate and, frankly, predictable — critics are right to point out that a camera on someone's face raises consent questions for everyone standing near that person, not just the wearer. Zuckerberg's dismissal of "AI doom" narratives reads less like a considered rebuttal and more like a founder who doesn't want existential-risk talk slowing down a product launch. That's a business incentive, not a safety argument, and it's worth keeping those two things separate.

Meanwhile, Amazon took the more mundane but arguably more consequential path this week, rolling out agentic AI tools for third-party marketplace sellers. No headlines about superintelligence, just automation aimed at inventory, listings, customer service — the unglamorous plumbing of e-commerce. This is where agentic AI is actually going to earn its keep in the near term: not in glasses that reimagine human perception, but in back-office tasks that free up small business owners to do literally anything else. I'd argue Amazon's move matters more for the average person's economic life than Meta's wearable, even though it will get a fraction of the coverage.

And then there's Spain, which reported what's being called the country's first data breach attributed to an autonomous AI agent. Details are still thin, but the symbolism is hard to miss — we've moved from theoretical warnings about agentic AI security risks to an actual documented incident. This lands the same week that Dario Amodei and Sam Altman both briefed the UN Security Council on AI safety, with Amodei warning that poorly managed AI could pose risks to humanity broadly. It's easy to roll your eyes at AI executives testifying about the dangers of the technology they're racing to build faster than their competitors — the incentive misalignment there is obvious. But the Spain case suggests the gap between "hypothetical agent risk" and "agent risk you have to file a police report about" is closing faster than most regulatory frameworks can track.

By the way, Brookings published a piece this week asking plainly whether the AI existential threat is real and whether regulation can actually stop it — a useful corrective to both the doom-mongers and the Zuckerberg-style dismissals. I don't think anyone has a satisfying answer yet. What I do think is that we're going to get a lot more clarity from incidents like Spain's breach than from another round of Security Council briefings.