How to — cut an AI app off from your accounts

Share
How to — cut an AI app off from your accounts

You connected an AI tool to your mail or calendar once, used it for a week, and stopped thinking about it. By the end of this, every connected-apps list you own will show only apps you can name a reason for — and you'll know exactly how to pull the plug.

What you granted wasn't a one-time peek. When you tapped "Allow" on a consent screen, you issued the app a standing key: a bundle of permissions, which the standards call a scope, that keeps working while you're not looking. That's the point of the design — an email assistant that could only read your inbox while you watched it would be useless — and it's also why access accumulates. Every product that ever talked you into connecting still holds whatever you handed it.

Detailed view of smartphone displaying multiple app icons on screen, highlighting technology use.

It matters more now than it did two years ago, because the tools holding these keys increasingly act on their own: drafting, sending, booking, filing. A forgotten grant is a door nobody is guarding.

1. Start on the side that granted the access

Deleting the AI app from your phone or closing its tab does nothing to the keys it already holds. The grant lives on the account you approved it from: Google's list of apps with access to your data, your Apple Account's Sign in with Apple apps, your email provider's delegated access, each social platform's connected-apps page. Walk every service where you've ever tapped "Connect" — people check the big account and forget the mail provider, the job board, the old portfolio site. This is also the case for why you care at all: What is prompt injection? explains how instructions get smuggled into tools an app can still reach — and access nobody checks is access nobody is watching.

2. Sort by blast radius, not by install date

Read what each grant actually permits, in plain terms: can it read or write? Can it send messages as you, or only read them? Does it work anytime, or only while you're watching it? The powers that matter are send, write, and manage — over mail, calendars, files, and payments. A chatbot that can read your notes is a smaller problem than a "personal assistant" that can send email from your address, no matter which one looks fancier. For the other half of the risk — what can go wrong while a connected app is quietly doing its job — How to — spot prompt injection in a product you use covers the warning signs.

3. Revoke anything you can't explain in five seconds

If you can't name the job an app still does for you, it doesn't keep the keys: the abandoned trial, the duplicate assistant, the expired job-search tool. Grants routinely outlive the products behind them. Companies fold without ever revoking their tokens, rebrands leave old permissions under old names, and the free tier you stopped opening in March still has exactly what you gave it in March. You are the only party to this relationship who remembers it exists.

4. Re-grant the keepers narrowly

Consent screens are worth reading when they reappear. Which account is being connected? Which powers, exactly? How long does the grant last? Give read access first and let the tool ask for write access when it has something real to write. When a product offers a masked or relayed email address and you don't expect mail from it, take the masked one — the app gets an identity, not a forwarding address into your inbox. This is the account-owner version of the question we asked builders in How to — decide what an AI agent may do: you don't have to be building agents to answer it for yourself.

5. Sweep the side doors

Revoking the headline grant doesn't clean up everything the app left behind. Mailbox delegates, app-specific passwords, forwarding or import rules, and old sessions inside the tool's own account all survive an OAuth revocation. And every connection has two sides: the account that issued the key and the app that holds it. Disconnect both — revoke at the source, then remove the listing inside the AI product itself, so it stops quietly asking for data you thought you cut off.

Don't do this

Don't revoke everything in one furious sitting. The calendar sync, the email helper, the meeting-notes bot you actually rely on — they all die at once, and you'll spend an afternoon debugging products that are merely locked out. Pull the grants one at a time, or set aside a re-authorization week and accept it as the cost of the sweep.

How you'll know it worked

The connected-apps lists now show only apps you can name a reason for. The practical test: pick one app you revoked and let it wake up. Instead of quietly doing its job, it should come back asking you to sign in again. If it keeps working, the grant didn't die where you pulled it — go find where it actually lives: a second provider, an app-specific password, or a session it never lost.

Which AI app still has access to your inbox that you haven't opened in a month? Tell us in the comments.

Read more

OpenAI drops 372 math results, nearly all from one prompt

OpenAI drops 372 math results, nearly all from one prompt

A dense news day across three fronts: OpenAI published a mountain of mathematics from an unreleased model, SpaceX reportedly wants the debt market to fund its Nvidia orders, and Anthropic restructured its cyber program around three tiers of trust. OpenAI published 372 mathematical result families from an internal frontier model — and says nearly all of them came from a single prompt handed to a single AI agent. The release comprises 722 manuscripts touching hundreds of open questions, plus Lea

Underdog launches a private on-device AI assistant, backed by a16z

Underdog launches a private on-device AI assistant, backed by a16z

The privacy split in consumer AI got a new entrant tonight, GitHub turned code-review benchmarks into a vendor-bias argument, and OpenAI opened its oddest API to everyone. Underdog launched in invite-only beta: an on-device AI assistant that keeps your data on your machine and never charges you a subscription. Self-taught coder and Thiel Fellow Sigil Wen — who moved to Silicon Valley at 17 and lived in an AI hacker house with Andrej Karpathy — built his own inference engine, Husky, to run a 2

Today in AI — October 6, 2026

Today in AI — October 6, 2026

A day of second-order moves: the labs are buying task data from software vendors instead of scraping the web, Waymo is borrowing to fund the robot world, and the biggest bank in the US just put a number on what Anthropic's latest model cost it in risk. Models & Research * OpenAI is training GPT-6 Astra on Ironclad's real contracting work. Ironclad staff helped turn 11 tasks across legal, commercial and procurement work — setting up NDAs, approval workflows, clauses that change by jurisdict

Meta, Walmart and Stripe publish the Personal Agent Protocol

Meta, Walmart and Stripe publish the Personal Agent Protocol

The agent economy is writing its rulebook tonight: one open standard for AI bots at the checkout, a cheaper image model from Google, and Anthropic turning bug-hunting into a tiered product. Meta, Walmart, Stripe and Sierra are publishing an open "personal agent protocol" — a standard that defines how personal AI agents interact with businesses online. The group behind it reads like a cross-section of agentic commerce: Meta, Sierra, Walmart, Stripe, Shopify, Genesys, Rocket, NiCE, Decagon and I