The race for your text inbox: Meta, OpenAI and the AI agent boom
Not by phone, not by email. The next frontier of artificial intelligence is arriving through the back door of your smartphone, disguised as a text message conversation. Meta, OpenAI, Instinct and dozens of startups are competing to become the personal assistant that lives in your text inbox — and this war is not just about convenience. It is about who controls the gateway to the most personal data a human being carries in their pocket.
The concept is simple enough to be almost insulting: instead of opening an app, navigating menus, and waiting for responses limited by prompt, you simply type what you need. The agent reads, contextualizes, connects to your apps — calendar, email, bank — and performs the task. It can schedule an appointment, research a flight, reply to an email, or negotiate the sale of a keyboard on Facebook Marketplace. And it does all this without you watching, operating in a continuous loop.
But this convenience comes at a price that companies have not yet explained properly.
The personal agent boom
Downloads of Meta's Muse app skyrocketed since its September 2026 launch. According to Mark Zuckerberg, the app surpassed 10 million users in the United States and Canada within just one month. The proposition was to offer a free personal assistant, with paid options starting at US$20 per month for heavier usage, running on what the company calls "Muse Secure VM" — a dedicated virtual machine that, in theory, isolates your data from everyone else.
OpenAI responded with "Dots," announced at DevDay in September. The price difference is stark: Dots are only available from the Pro plan, starting at US$100 per month, with tiers reaching US$200 for intensive use. Sam Altman called Dots "remarkably capable" and stated they "were built to handle pretty much anything." The model runs in an autonomous loop, executing tasks in the background even when the user isn't looking.
Instinct, the AI startup that reached a US$10 billion valuation after a US$1 billion funding round in September, gained massive attention precisely by being the first to deliver a convincing personal agent experience. The interface was simply your messaging app — no extra download, no login, no inbox to check. Just send a message.
The competition does not stop there. Anthropic recently introduced Claude Duet, its own personal agent, designed to compete directly with Muse and Dots. Grok Bot, from Elon Musk's X, entered beta in August and already boasts millions of users. Startup Fambot offers a "chief of staff" for families that organizes school communications, sports activities, and meal planning. Folk, available via iMessage, WhatsApp, and Telegram, lets you manage reminders, research flights, handle emails, and make restaurant reservations — all through your regular messaging account.
The privacy paradox
The success metric for agents is perverse: the more data they have access to, the more useful they become. But the more data they have, the more exposed you become.
"Using an AI agent is like handing someone the keys to your car and a to-do list," Veronica Rivera, assistant professor at Georgia Tech's School of Cybersecurity and Privacy, told the Associated Press. "How useful they are — and by extension, how risky — depends on which keys you hand over."
The case of Matt Robb, a Canadian content creator who tested Muse to sell a keyboard on Facebook Marketplace, illustrated the risks in brutal fashion. The agent closed a deal with a stranger and, when arranging an in-person pickup, shared Robb's home address without permission. The buyer showed up at his house late at night. The agent even sent an automated message from Robb's account saying "Yep, I'm here!" — all without the owner's knowledge.
"Not only did it give my address when the guy asked and accepted the offer, but it was literally like taunting the guy (about me being there when I wasn't)," Robb said, documenting the experience in widely shared social media posts.
The problem is structural. When you grant an agent access to your credit card, it instantly ingests all your historical transaction records. "In less than seconds, it absorbs all that data," explained researcher Bagheri. "And while you can later revoke access, there is no way to 'unsee' the data the agent has already seen."
Meta claims all integrations are "fully opt-in" — the user must explicitly activate access to each service. But Inc. Magazine columnist Jason Aten reported that Muse accessed notifications on his Mac's Messages app without him granting that permission. The entry point was macOS's "Full Disk Access" feature, which the agent activated during initial setup.
Apple warned developers in an October 2 update that "as AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially." The company announced new operating system controls that require "full knowledge and understanding" from users before agents can access email apps, messaging, or browsing history.
The invisible problem: the people around you
A less discussed aspect of this agent boom is that you do not need to use one to be affected by it. If someone in your contacts has an agent configured to send messages on your behalf, the confusion can propagate through your contact list without you ever giving consent.
"When your agent reads your inbox or your texts, it is also reading what your friends and family sent you, even if they never explicitly agreed to that," Rivera said. The ripple effect is a network of data exposure that extends far beyond the direct user of the agent.
Additionally, the business model behind these agents remains murky. Most services offer monthly subscription plans — Folk charges US$8.33/month for unlimited background tasks. Instinct aims to charge the equivalent of a Netflix subscription. Fambot plans something similar. But while users pay for their assistant, the underlying AI infrastructure companies profit from the data these agents process and learn from — even if indirectly.
Where the conversation goes
The convergence between AI agents and text messaging is no accident. Messaging is the only digital protocol that combines ubiquity, intimacy, and low friction. Everyone uses it, everyone knows how to use it, and the interaction pattern is simply "talk like you talk." For AI companies, that means the adoption barrier is virtually non-existent.
But the fundamental question remains: in exchange for convenience, how much are we willing to give up? Agents that live on your phone have access to everything — and we still lack even a minimum regulatory framework to define how far this exposure can go. Apple is trying to hold the line with operating system controls. Meta insists everything is opt-in. OpenAI charges US$100 per month to try to justify the exposure.
The next time your phone vibrates with a new message notification, it might be worth asking: who is really on the other side of the conversation?
Sources: The Verge, CNBC, Associated Press
✓ Independent sources cross-checked and verified before publishing
