AI assistants are beginning to do much more than answer questions, summarize documents, or help write an email. Some can connect to other applications, gather information across systems, and take actions on your behalf.
I recently discovered Instinct, and it made me think more seriously about what changes when AI moves from helping us prepare something to being able to do something for us.
Instinct describes itself as a personal assistant that can connect to applications and devices including email, messaging, your screen, audio, location, and more. It is designed to handle tasks such as following up on conversations, arranging transportation, and coordinating services.
That creates a different question from simply asking, “Can AI write this for me?” As these tools become more capable, we also need to ask: What should AI be allowed to do without me?
AI privacy does not have to be all or nothing
One reaction to increasingly capable AI assistants is to avoid them because you do not want to hand over your email, passwords, financial information, private documents, or personal data. Those concerns deserve to be taken seriously, especially with tools that can access other services and act on your behalf.
But you do not have to begin by connecting everything.
There are plenty of useful things you can do with an AI assistant using public information, creative ideas, hypothetical situations, public business information, and limited details that are proportionate to the task. That gives you a chance to learn what the tool is good at before deciding whether deeper access is worth it.
16 Ways to Try Instinct Without Adding Sensitive Personal Information
1. Research speakers before a conference
Give Instinct the public conference website and ask it to create a briefing on the speakers. It can summarize their backgrounds, areas of expertise, recent interviews, notable ideas, and useful questions you might ask if you meet them.
Everything can come from public sources, so there is no need to connect your contacts, messages, or other personal accounts.
2. Organize a conference schedule
Give it the public event agenda and ask it to organize sessions by day, time, topic, or priority. You could also ask it to identify overlapping sessions or recommend which ones best match what you want to learn.
You can get a surprisingly useful conference plan without giving the AI access to your personal calendar.
3. Research a course before buying it
Give Instinct the public sales page for a class, coaching program, or training course. Ask it to examine the curriculum, pricing, instructor background, refund policy, what is included, and independent reviews.
Then ask it to identify anything that is still unclear before you spend the money. You can do the research without providing a credit card or payment information.
4. Look for a less expensive way to access something
Before paying for a conference, webinar, training program, subscription, or other offer, ask the AI to search public FAQs, organizer announcements, previous offers, replay options, pricing pages, and other public information.
You might discover a lower-cost tier, replay package, free trial, previous promotion, or alternative way to get what you need.
5. Draft the questions you should ask before buying
Sometimes public research does not answer everything. Ask the AI to prepare a concise inquiry covering pricing, refundability, deadlines, access length, downloads, cancellation terms, or anything else you still need to know.
You can review the draft yourself and send it through your own email account.
6. Research a travel problem
Give it information such as the airline, route, flight number, airport, or travel date and ask it to research delays, alternate routes, airline policies, schedules, or other options.
You do not necessarily need to provide your passport, credit card, airline password, or identity documents to begin understanding your choices.
7. Study another company’s marketing strategy
Give it a public website, campaign, landing page, webinar, social account, or article and ask it to analyze what the company appears to be doing.
Who are they targeting? What are they selling? What is the offer? How are they positioning it? What marketing techniques are being used? What could a small business learn from the approach?
This is one of the areas where AI can become more useful than simply summarizing a webpage.
8. Turn an idea into an advertisement
Describe a fictional character, business scenario, product idea, or marketing concept and ask the AI to develop scripts, headlines, campaign directions, or a short advertisement.
Creative experimentation does not require customer records or private business information.
9. Turn the same idea into another format
Take an idea you already developed and ask the AI to turn it into an Instagram carousel, comic, short video concept, presentation, infographic, or another format.
This is also a useful way to test how well an AI tool understands an idea rather than simply generating something from scratch.
10. Build a fact-based content piece
Give it a topic and ask it to research studies, statistics, reporting, arguments, counterarguments, and original sources. You can then use that research to build an article, presentation, newsletter, workshop, or educational piece.
For important subjects, I would still open the original sources myself rather than relying only on the AI’s summary.
11. Find audiences you may not have considered
Describe a product, service, event, campaign, or idea and ask who else might care about it. The AI may identify different customer groups, partners, professional associations, communities, decision-makers, or distribution channels you had not considered.
You can start this exercise without giving it your customer database.
12. Pressure-test an idea
Give the AI your idea and ask it to challenge you. Ask what assumptions you are making, what could go wrong, what a customer might object to, what competing options exist, and what information someone would need before saying yes.
I like this use of AI because the value comes from improving the thinking behind the idea rather than simply producing more content.
13. Apply a public business framework to a hypothetical company
If you find an interesting strategy, framework, article, or business model, ask Instinct to apply it to a fictional restaurant, salon, consulting business, retail store, or another example.
You can understand how the framework works before deciding whether you ever want to provide your own financial statements, customer information, payroll data, contracts, or internal documents.
14. Extract and organize public information
Give it a public report, directory, guide, resource page, or long list and ask it to categorize the information, identify patterns, summarize it, or turn it into a more useful working reference.
This can be especially useful when the information already exists but is scattered or difficult to work with.
15. Fact-check something you see online
Ask the AI to investigate a viral claim, news story, statistic, quote, or social media post. Have it distinguish what is confirmed, what is disputed, what remains unknown, and where the original information came from.
The AI can accelerate the research, while you retain responsibility for checking the important sources yourself.
16. Ask Instinct to investigate Instinct
Ask the tool to research criticism of itself, review its privacy policies and terms, find security reporting, identify potential risks, and explain what safeguards a user might consider.
This could be one of the better first experiments because you are learning about the tool while also seeing how it handles research that may be critical of the product itself.
Notice what you did not have to provide
None of those examples inherently requires your password, bank account number, credit card information, private client documents, health records, confidential contracts, tax documents, or unrestricted access to your inbox.
Some tasks may benefit from limited personal information. Researching a specific delayed flight, for example, could require a flight number and destination to be useful.
The better question is whether the information you are sharing is proportionate to the task you are asking the AI to perform.
“AI can create privacy risk” does not mean “every useful AI task requires sensitive data.” A lot of value can come from a public link, a public question, a creative idea, a hypothetical scenario, or ordinary business context.
Want to experiment with Instinct?
If you want to see the tool that prompted me to think more about this, you can try it with my invite link:
I would start with one of the lower-risk experiments above before connecting anything sensitive. Give it a public website, ask it to research something you are considering buying, have it pressure-test an idea, or give it a public event agenda and see what it can do.
Then decide whether additional access would solve a problem important enough to justify it.
What changes when you give AI more access?
This is where my rule becomes:
Broad preparation. Narrow execution.
AI can have broad freedom to prepare work while you maintain tighter control over actions that create commitments or consequences.
| Broad preparation | Narrow execution |
|---|---|
| Research | Send |
| Compare | Publish |
| Organize | Purchase |
| Summarize | Pay |
| Analyze | Book |
| Calculate | Cancel |
| Draft | Delete |
| Find options | Share externally |
| Prepare recommendations | Change permissions |
| Identify questions | Enter commitments |
The difference matters with an agent like Instinct. Its terms say the service can access and interact with connected services, exchange data with them, and take actions on a user’s behalf. They also explicitly contemplate purchases and even agreements, commitments, or transactions made through those actions. (Instinct)
The same terms warn that AI actions can contain errors, may not always be reversible, and should be independently verified. They also say Instinct may implement safeguards or confirmation requirements but does not guarantee those safeguards will prevent unintended actions. (Instinct)
That leads to the guardrails I would use with Instinct or any other AI agent.
My 15 Rules for Giving AI More Access
1. Require exact approval before anything leaves your account
Let AI prepare the email, social post, invitation, document, or message. Before anything goes out, review exactly what is being sent and who will receive it.
Drafting and sending should be treated as two different steps.
2. Require exact approval before money moves
Before an AI purchases or books something, check the item, total cost, payment method, address, cancellation rules, and refundability.
A recommendation to purchase something is very different from permission to purchase it.
3. Keep destructive actions behind a review step
Deleting a file, canceling a reservation, ending a subscription, removing a calendar event, or closing an account deserves additional scrutiny.
Do not approve an irreversible action based only on an AI-generated summary. Look at the item itself before approving the action.
4. Treat outside content as information, not authority
An AI agent may encounter instructions while reading an email, website, document, or message. Those instructions came from someone else and should not automatically become instructions from you.
The more systems an AI can interact with, the more important that distinction becomes.
5. Give important instructions through the AI system you control
An email sender, website, document, or another AI should not be able to claim that it speaks for you. Your authenticated interaction with the AI should remain the source of authority for important actions.
Think of outside material as something the AI can read, not something that automatically gets to tell it what to do.
6. Keep persistent secrets out of ordinary conversations
Passwords, recovery codes, API keys, private keys, PINs, authentication cookies, and similar credentials deserve stronger protection than ordinary information.
Do not paste them into an AI conversation simply because the AI asks for them.
7. Connect only accounts that create real value
Every integration expands what an AI may be able to see or do. If you never need an AI to interact with a particular account, there may be little reason to connect it.
More access should solve a real problem, not simply be enabled because the option exists.
8. Review your connected accounts regularly
Periodically check which email accounts, calendars, cloud drives, applications, devices, and other systems remain connected.
Remove connections that no longer provide enough value to justify the access, especially as you experiment with more AI products over time.
9. Understand that disconnecting is not necessarily deleting
This is easy to overlook. Instinct’s current terms say that even after disconnecting a connected service, indexed data from that service may remain unless you separately request deletion. (Instinct)
That means it is worth understanding both how to disconnect an integration and how to remove information that has already been collected.
10. Use the narrowest useful permission
If reading information solves the problem, think carefully before granting permission to edit, send, purchase, or delete.
If one account is sufficient, do not connect three. Give the AI enough access to accomplish the job without automatically giving it every permission available.
11. Keep financial access minimal
Do not store a payment method or financial credential simply because you might want an AI to make a purchase someday.
Add sensitive access when there is a clear recurring use case and the value is high enough to justify it.
12. Pay attention to browser access
Formal integrations are not the only source of access. A browser may already be signed into email, ecommerce platforms, social accounts, cloud software, banking sites, or other services.
Review stale authenticated sessions and sign out of accounts an AI-enabled browser does not need.
13. Review how your information may be used
Privacy and model-training practices vary between services, so do not assume every AI product handles your information the same way.
Instinct’s current terms say users can opt out of future use of their materials for AI model training, subject to some exceptions, and that the change applies going forward rather than undoing prior training.
Settings and policies can change, so this is something worth reviewing periodically rather than setting once and forgetting.
14. Separate preparation from execution
Let AI research the vendors, compare the flights, organize the options, draft the email, prepare the shopping list, or analyze the choices.
Then create a deliberate checkpoint before the action becomes a commitment.
You may eventually decide that some routine execution tasks are safe enough to automate. That decision should be intentional rather than the default simply because the technology is capable of doing it.
15. Verify the outcome after every sensitive action
An AI reporting that a task succeeded is not the same thing as confirming that the real-world result matches what you intended.
Instinct’s terms specifically tell users to independently verify actions and note that records of actions available through the service may not always be accurate.
Here is what I would check:
| If AI did this | Verify this |
|---|---|
| Sent an email | Recipient, content and attachments |
| Made a purchase | Product, quantity, total and payment method |
| Booked travel | Date, route, passenger, seat and fare conditions |
| Scheduled a meeting | Date, time, timezone and attendees |
| Made a reservation | Location, date, party size and cancellation terms |
| Canceled something | The correct reservation, subscription or event |
| Shared a file | Document, recipient and permissions |
| Deleted something | The correct file, record or account |
How much access does AI really need?
Before granting another permission, I would ask four questions.
| Question | Why it matters |
|---|---|
| Does the task require this information? | If it does not, there is little reason to provide it. |
| Could read-only access accomplish the job? | More permission does not automatically create more value. |
| What happens if the AI gets this wrong? | The consequence should determine how much human review you require. |
| Is the time saved worth the additional access? | Convenience alone may not justify a significant increase in access. |
This is a more useful way to think about AI privacy than simply trying to decide whether a particular tool is “safe” or “unsafe.”
The same AI assistant can present very different levels of risk depending on what you ask it to do. Researching public conference speakers is a very different situation from connecting email, payments, calendars, private business documents, and customer information.
The question businesses need to start asking
For the last few years, a lot of the AI conversation has centered on capability. We have asked whether AI can write something, analyze data, make an image, research a topic, or summarize a document.
Agentic AI introduces a different question: Should AI be allowed to do this without me?
For a small business owner, that distinction can become particularly important. One account may contain customer conversations, contracts, financial documents, calendars, vendor information, marketing plans, and access to other business systems.
The productivity upside can be significant, but so can the consequences of an action going wrong.
That is why I keep coming back to the same principle:
Broad preparation. Narrow execution.
Give AI plenty of room to help you research, think, organize, compare, draft, and prepare. Be much more deliberate when an action affects your money, reputation, customers, access, commitments, or something difficult to reverse.
You do not have to choose between giving AI access to everything and refusing to use these tools at all. There is a very large and useful space in between, and that is a good place to start.
What are you asking AI to do today that gives you useful help without requiring sensitive personal information?