If you have not yet received a message from an AI agent that knows your calendar, the rest of this article is your early warning. In a round-up first published October 3, 2026 by Lauren Forristal at TechCrunch (since updated), at least fifteen vendors now ship AI assistants designed to be reached the same way you reach a person: by SMS, iMessage, WhatsApp, or Telegram. One, Instinct, has raised $1 billion at a $10 billion valuation; others report seed rounds of a few million dollars. One is already on Apple’s Messages for Business platform. The privacy model is, in most cases, whatever the vendor’s defaults allow, and those defaults are still moving.
What an SMS-resident agent actually reads
The first thing to understand about this category is that the agent is not a chatbot. According to Forristal’s survey, the agents in this class take actions on the user’s behalf: scheduling appointments, organizing calendars, sending email, making reservations, shopping online, and in some cases signing up for services or placing calls. What they can touch is whatever the user connects to them, and whatever third-party service the agent is permitted to act on.
A partial list of what these agents connect to, drawn from the same TechCrunch piece: calendars, Gmail inboxes, Google Workspace, Google Drive files, WhatsApp, Slack, and a user’s Instagram account. Some go the other way and give the agent its own identity: Wajo’s agent Fo has its own email address, phone number, and payment card. Caddy lives in iMessage and RCS and connects to calendars and conversations to find things that need action. Fambot reads Gmail and Google Calendar to produce nightly next-day summaries. Folk runs on a private cloud computer and can execute multi-step tasks. Pally bridges WhatsApp, Gmail, calendars, and Google Drive. Instinct has begun giving each user a dedicated email address the assistant can use to register for services.
That is the practical reason an SMS-resident agent is a different privacy object from a chatbot on a webpage. A chatbot is a perimeter. An SMS agent is a participant. It sits inside the most intimate channel on the phone, alongside one-time codes, two-factor prompts, and the people you actually know.
Poke and Apple’s door
The category got its first platform-level stamp of approval in June 2026, when Apple let Poke onto its Messages for Business platform. Until that approval, Messages for Business was designed for businesses such as airlines, retailers, and hotel chains to communicate with their own customers, and had not been open to stand-alone third-party AI agents. Poke, built by the Interaction Company of California, is the first standalone AI agent to clear Apple’s review for that surface.
In the same June 4, 2026 TechCrunch report by Sarah Perez, Poke co-founder Marvin von Hagen said, “I think Apple is just noticing this is the best way to offer AI.” The company had to verify it could offer live support, clearly disclose that the conversation partner is automated, customize its UI to Apple’s style guide, and submit testimony from its messaging providers. The Interaction Company of California was reported as a 10-person team at the time, backed by Spark Capital and General Catalyst. According to the October round-up, it was acquired a month later by AI coding startup Cognition in a deal valued in the low nine figures.
The significance is what Apple’s platform now permits. Messages for Business is a verified business channel inside the Messages app, so a user can text an AI agent there the way they would text a company, without downloading a separate app. Apple’s approval puts a third-party agent in that channel for the first time.
Instinct, money, and a privacy policy reset
If Poke is the platform story, Instinct is the money story, and the cautionary one. The October 3 round-up noted that Instinct’s level of autonomy had drawn privacy and security criticism. The same company announced a $1 billion Series C at a $10 billion valuation on September 28, 2026, per TechCrunch’s reporting, about a month after a fundraise that valued it at $2.5 billion. Investors include Sequoia Capital, Benchmark Capital, and Coatue. Instinct launched as an invite-only service in August 2026 and uses its own phone number and computer to perform tasks, with features that include a “concierge” that makes phone calls and a “trusted person network” that lets one person’s agent coordinate plans with friends’ agents.
The privacy flag comes from the September 28 funding story, which called the initial version of Instinct’s privacy policy “particularly worrisome due to its overreach” and noted that the policy has since been updated. The combination - a $10 billion valuation, an agent that can make calls and act on your behalf, and a privacy policy that had to be revised - is a telling data point in this category so far on the gap between what an SMS agent can do and what its users have consented to.
What This Means
The privacy question for an SMS agent is not the same as the privacy question for a chatbot, and the difference is consent surface. A chatbot is a thing you visit. An SMS agent is a thing you give your phone number to. Once the agent is in your thread, its vendor sees everything you send it and when. If the vendor is also granted access to Gmail, calendar, contacts, or payments, the consent surface expands sharply, and most onboarding flows do not slow down to ask the user to read the new permissions line by line.
Three concrete implications follow. First, Instinct’s revision of a privacy policy criticized for overreach is the closest thing this category has to a consent precedent. Future SMS agents will be expected to ship with policy copy that does not require revision. Second, Apple’s Messages for Business approval sets a quality bar: live support, explicit AI labeling, style-guide compliance. Von Hagen said any other agent seeking the same approval should expect a couple of months to get through it. Third, the operators of these agents hold a security surface that did not exist when the only AI in your phone was a keyboard autocomplete: the agent now holds tokens for email, calendar, drive, and payments. A compromise of the vendor is a compromise of your inbox.
The practical action for a reader is the same as it has been for every consumer AI product since 2024: read the onboarding, count the permissions, assume anything you give the agent will eventually be read by a human during an incident review, and turn off anything you would not want a stranger to see. The action for the industry is harder. It requires deciding whether the consent defaults in this new surface deserve a $10 billion valuation, or whether they need to be rewritten first.
The Bottom Line
At least fifteen SMS- and messaging-resident AI agents are now available or in beta, one of them is the first third-party AI agent on Apple’s Messages for Business channel, and the category’s highest-funded entrant has already had to revise a privacy policy criticized for overreach. The agent in your text messages is the new perimeter. The perimeter is still being defined.
Related on Intelligibberish
- FTC’s First AI Agent Probe Names OpenAI, Anthropic - the first regulatory test case for the same category of agent.
- How to Delete Your Data From ChatGPT, Claude, and Gemini - the practical companion when an SMS agent has read more than you wanted.
- Privacy - more on consent surfaces, data retention, and what defaults actually allow.