Is the Hark AI assistant more private than Muse or Instinct?

What did Hark just launch, and why is privacy its pitch?

Hark launched Hark Pro on October 6, 2026. It's a personal AI assistant that runs errands on the web for you, and the company is selling it mainly on privacy. Hark is the startup led by Figure AI founder Brett Adcock. The assistant works on the web, iOS and Android. There's a free tier plus paid plans at $20 and $100 a month, and the first 100,000 people to sign up get the paid plan free. Underneath sits an agent called Handoff. For each task it starts a separate virtual computer with its own browser, so it can order food, shop, book flights and pay bills. Hark calls the whole thing an operating system for your life. That's a big claim for a product that's a day old, but the ambition is clear.

Hark is pitching privacy because the market is suddenly crowded with assistants that want access to everything. Meta released Muse in September. OpenAI unveiled Dots on September 29. Instinct has been running an invite-only beta since February and has drawn a steady stream of privacy questions. Against that crowd, Hark makes a four-part pledge: your data is yours, it will never sell it, it will never share it with advertisers, and you can always delete it. Passwords and payment cards go into an encrypted vault the company says even it can't open. Hark says the outside AI models it uses keep zero data. While the agent works, a small live window shows it moving through websites, so you can watch instead of guessing.

For anyone who runs a business from their phone, the question has changed. It used to be which assistant is smartest. Now it's which one you'd trust with your inbox, calendar, client documents and company card. Here's an opinion, and we'll defend it: a privacy pledge on a launch page is cheap, and defaults are what count. What does the assistant reach on day one? Does it ask before it spends money? What happens to your data after you disconnect it? This article compares Hark, Muse, Instinct and Dots on those three things. It ends with a checklist you can use on any assistant, including the ones that live inside WhatsApp.

What can Hark, Muse, Instinct and Dots actually access?

All four assistants can reach deep into your life: email, calendar and files, and most of them can also pay for things. What separates them is how far that access goes. Instinct reaches furthest. Its privacy notice says it may access what's on your screen, your screenshots, cursor movements, keyboard input, messages and emails, and audio and voice when you turn that on. It can also use your location. You hand it logins and a payment card so it can sign in and finish tasks. That's how it books travel, buys groceries, cancels subscriptions and sends email for you. It's impressive, and it means one company holds nearly every way into your accounts.

Meta Muse goes deep in a different way: memory. Muse is built to remember what matters to you, including details about the people you mention. Researchers found it builds detailed profiles of users' friends and family. It reviews the day's conversations every night and tries to work out goals you haven't said out loud. With your permission it can read your email and other common apps, and you pick which apps it connects to. Meta says Muse data won't feed its ad systems, and that Muse can't see your passwords or payment methods. The catch is other people. Your clients never agreed to be profiled, but they still are once your inbox is plugged in.

OpenAI Dots and Hark Pro sit closer together. A Dot is a separate computer in OpenAI's data centers. It keeps working whether or not you're looking, connects to more than 4,000 apps including Slack and Teams, and you can even phone it. Hark Pro reaches email, calendars, files and payments to finish tasks from start to end, and it keeps card details in its vault. One point causes confusion. Some early write-ups said Hark runs on your device and never uploads anything. Most reporting describes the opposite: tasks run on cloud computers, and the app is just your window onto them. Treat that as true for all four. None of these assistants really lives on your phone.

Which AI assistants act without asking you first?

Hark asks before it acts by default. Instinct and Dots are built to work more on their own. Hark's security page says it starts every task by defaulting to consent. In one public demo, Hark's head of design ordered a sandwich through a delivery app, and Handoff still stopped for a final yes before paying. The catch is that Hark remembers the actions you're happy to hand over. Its terms let you give standing permission for a whole category of purchases, and after that each purchase in the category goes through without a prompt. Reviewers flagged that clause straight away, because it's how a helpful habit turns into charges on your card.

Instinct goes further. Its terms say you authorize it to take actions on your behalf as it deems responsive to your input. In plain English, it decides what your request means, and what it agrees to is binding on you. Its own privacy policy warns about unintended payments and about hidden instructions in web pages or emails that could steer the agent. Dots is designed to chase the goals you set, continuously, in the background, with minimal oversight, and OpenAI talks about teams of Dots working together for you. Muse also works on its own cloud computer. Meta tested sending some of its phone calls to human contractors, then paused that test over privacy concerns.

This is where we disagree with most of the launch coverage. Encrypted vaults get the headlines, but the confirmation step protects you more. A vault keeps your card number safe from hackers. It does nothing when the agent itself misreads 'book something near the office' and pays for the wrong hotel. An assistant that shows you exactly what it's about to do and waits for a yes makes it much harder for a mistake to go unnoticed. So whatever you pick, turn standing permissions off or keep them narrow, at least for the first month. Start with small tasks you can undo, like a cheap grocery order or a restaurant booking, and note every time it asks you to confirm.

What happens to your data after you disconnect or delete it?

This is where the assistants differ most, and it's the question almost nobody asks before signing up. Hark promises you can always delete your data. Instinct's terms say that disconnecting an account does not delete the data it has already indexed from it. One early tester disconnected their email and found Instinct still summarizing their inbox. When they asked why, the assistant said the emails were stored in plain text for later searches. On top of that, the terms give the company a perpetual, irrevocable licence to use your material to run, develop and train its services. You can do little to unwind that once you've connected everything.

Meta takes a middle path, but the defaults lean towards Meta. Your Muse conversations are used to train Meta's AI unless you switch that off. Meta says Muse data stays out of its ad systems, and you can opt out of training. A stronger option has been announced but hasn't launched yet: an encrypted cloud computer that only you hold the key to. Dots is too new for anyone to know how OpenAI handles retention over months of real use. It's only open to paying Pro and Business Premium customers, and no business has run it long enough to report back. Assume nothing. Before connecting a single account, read the data and training settings.

Whichever assistant you try, a few habits limit the damage if something goes wrong. Turn off model training before you connect anything, not after, because whatever it has already learned from usually can't be pulled back out. Give it a separate payment card with a low limit, never your main business account. Connect one account at a time and watch what it does for a week before adding another. Keep a note of everything you connected, so you know what to revoke if you leave. And ask the assistant directly what it remembers about you. A trustworthy one will tell you plainly. A vague answer tells you something too.

How do you judge any AI assistant's privacy, including WhatsApp ones?

Ask six questions before you trust any assistant: what can it reach, where does it run, does it ask before acting, who does it pretend to be, does it train on your data, and what happens when you leave. Run Hark, Muse, Instinct and Dots through that list and they split cleanly. Hark scores well on consent and pledges, Muse loses points on training defaults and on profiling people who never signed up, Instinct takes everything and keeps it, and Dots is too new to judge. The list works just as well for smaller assistants, including the ones that skip the app entirely and live inside a chat you already use every day.

Take Agent Sonic, which you message in WhatsApp like any other contact, with nothing to install and no dashboard. It isn't an agent that takes over a computer. It can't watch your screen, log your keystrokes or follow your location. It works with what you send it, like voice notes, PDFs, invoices and contracts, plus any Google Calendar, Sheets or Docs you connect, and it can read a Notion knowledge base without changing it. It also asks before it acts. Say 'Ask Dana when the invoice will be ready and let me know', and Sonic shows you the exact first message before anything goes out. Reply 'yes' to send it, 'no' to drop it, or 'shorter, no emoji' to change it.

Sonic also scores well on the 'who does it pretend to be' question. Outreach tasks, a Pro plan feature still in Beta, send from Sonic's own WhatsApp number, and Sonic always introduces itself as your assistant. It never poses as you or sends from your number. The limits are strict. You get one task at a time and up to three people a day, and first messages only go out between 08:00 and 21:00 in the other person's time zone. Anyone who says 'stop' is never contacted again. The trade-off is real: Sonic won't buy your groceries, and that's partly why it's easier to trust. If that sounds like the right balance, leave your details to get Sonic's number and start with 'Remind me to call the accountant every Monday at 9'.

Frequently asked questions