Hark releases an AI personal assistant with a focus on privacy
Oct 6, 2026, 11:22 AM · TechCrunch

Hark wants access to your email, files and credit cards so its agent can run your computer, and its privacy pitch rests more on trust in the company than on anything shown yet.
Why it matters
Hark, a startup founded less than a year ago by serial founder Brett Adcock, has widely released Hark Pro, a full-screen AI assistant built on a model trained specifically for computer use. It is free, with a paid tier for heavy users. Like rivals, it asks for access to email, calendar, hard drive and credit cards, then acts on a user's behalf.
Hark's design lead, former Apple staffer Abidur Chowdhury, pitches the company as the alternative to Meta and OpenAI, saying it is not there to sell ads and take user data. The company also promises an AI-native device in 2027.
From the desk
We think agents that handle real chores are worth building, and Hark has one smart idea we'd like everyone to copy. It shows a small window of the agent actually navigating the web. Most assistants hide that. Letting people watch the work is a simple, honest way to build trust, and it gives users a chance to catch a mistake before it costs them money.
The privacy promise is another matter. In the reporting we have, the case is a statement of intent: no ads, task-focused understanding of the user. That's welcome, but it isn't an architecture. We didn't see details on where data is stored, what stays on the device, how long it is kept, or what happens if the business model changes. A startup with no ads today is not a guarantee of no ads tomorrow. When the product holds your inbox, your files and your card numbers, those specifics are the whole story.
The broader risk is the one every computer-use agent shares. Software that can click, type and pay on your behalf is exactly what attackers want to trick or hijack, and the more of someone's digital life it touches, the bigger the blast radius. The anecdote of the assistant renewing a colleague's DMV registration is impressive. It is also a reminder of how much authority these systems are quietly being given.
There's an honest product question in here too. TechCrunch notes that people often struggle to think of useful tasks to hand off, and that a model trained narrowly for computer use may be faster and cheaper while giving up some broader knowledge. Our read: promising design instincts, a reasonable bet on a narrow model, and a privacy claim that needs to be backed with technical commitments before anyone hands over the keys.
Context
Hark is entering a crowded field of AI personal assistants that TechCrunch lists as including Muse, Instinct and Dots. Its stated aim is an operating system for future AI computers rather than a single app.
Who feels it
- Consumers
- A free agent that can do real tasks, in exchange for very broad access to accounts and payment methods.
- Competing assistant makers
- Hark's visible-agent window sets a transparency bar others may be pushed to match.
- Security teams
- Each new computer-use agent with financial access is another target for manipulation and account takeover.
What to watch
- A published privacy policy and technical details on data storage and retention
- Independent comparisons with Muse and other assistants on end-to-end task completion
- Security testing of Hark Pro against malicious websites and instructions
- What form the promised 2027 device takes