access token

advancedwidespread

An access token is a temporary digital entry card: a string of characters that allows a program to access your data without your password, for a limited time and with restricted permissions.

In plain terms

Think of a hotel room key card. You verify yourself once at the reception, and afterwards you just swipe the card: it opens your room and perhaps the wellness area, but not other rooms, and on the day of checkout it automatically loses its validity. The access token is the same in the digital world: a long, seemingly random string of characters that a program receives after your login and presents with every request.

It is useful so you do not have to give your password to every application. If a calendar app or an AI agent wants to access your emails, it does not get your password, but a token that authorizes only specific actions, for example reading, but not deleting.

It differs from a password in three ways: it lives for a short time (often only minutes or hours), has a narrow scope of authority, and can be revoked separately without changing your password. If it is lost, the damage is smaller, but while it is valid, it must be guarded just as carefully.

How it works

You log in with the provider (for example, Google) and approve what an application can access. This is the screen stating that 'the application would like to view your calendar'.
The provider then issues a token and hands it to the application. The token includes who owns it, what it authorises, and how long it is valid.
With every request, the application sends the token to the server holding your data, much like you swipe your card at a door.
The server checks whether the token is valid and whether the request falls within its authority. If yes, it fulfills the request; if not, it rejects it.
When a token expires, the application requests a new one with a longer-lived refresh token, or you must log in again. You can revoke access at any time in your account settings.

An everyday example

On a new website, you click the 'Sign in with Google' button and approve the site seeing your name and email address. The site then receives not your password but tokens: an identity token that verifies who you are, and an access token that authorizes it to retrieve exactly that much data. The same happens when you connect an AI assistant to your calendar or storage.

Who is affected

developerOn a daily basis, they authenticate their programs with tokens, and it is their responsibility to make sure these do not end up in public code or log files. → Code

everyoneWhenever you connect two applications or log in elsewhere with an account, you issue a token, and it is worth knowing that this can be revoked.

entrepreneur, executiveTokens for applications and agents accessing company data are assets as protected as passwords, which is why records and rules are needed for them.

bank customer, investorWhen you grant a financial application permission to view your account, the bank issues it a token, not your online banking password. → Money

A common misconception

MythThe access token is the same as the token in which AI models measure and bill text.

In realityThe two share only a name. A language model token is a piece of text (a word, word fragment or punctuation mark), while an access token is an access pass that proves authorization.

Limits and risks

Whoever obtains your token can act in your name until it expires or is revoked, so never paste it into a chat, email, or public code.
A token is only as secure as its scope is narrow: if an application requests access to everything and you grant it, it will have access to everything.
An AI agent can do anything the token allows with the entrusted token, even if it misunderstands your request or is deceived by malicious text.
Access granted to old, forgotten applications can persist for years, so periodically review your account settings and revoke what is no longer needed.

Similar, but not the same

password – You create the password, it opens your entire account, and it lives until you change it, while the token is machine-issued, has a narrow scope, and expires.
API key – An API key is typically long-lived and identifies the application or the subscriber itself, while an access token is short-lived and tied to a specific user's specific permission.
token (nyelvi modellben) – A language model's token is a text processing unit; it has nothing to do with authentication or permissions.

Related concepts

Broader

authentication, authorization management, cybersecurity

Narrower

refresh token, JSON Web Token, bearer token

On the edge

API key, OAuth, API, two-factor authentication, single sign-on

A brief history

Access tokens were already used by operating systems and network authentication earlier, but the web-based token as understood today was made common by OAuth. The first OAuth version was completed in 2007, and the OAuth 2.0 framework was published by the IETF in October 2012 under RFC 6749. The goal was that applications would not need to ask for users' passwords. OAuth 2.1 consolidates established security practices.

Where things stand

In late 2026, almost all web and mobile services handle access with tokens. The current question is how AI agents can obtain narrowly scoped, short-lived and easily traceable tokens. Protocols that connect agents to external tools, such as MCP, also build on OAuth for this.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

WorkOS, Northflank and Trigger.dev breached in attacks involving stolen AI access tokensNov 4, 2026 · Code

Also known as

access token · access token · access key · bearer token · OAuth-token · API-token · personal access token (PAT)

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.