Proof of identity and entitlement
Credentials are the information a person, device, or application uses to establish identity and entitlement to a service. Familiar examples are usernames and passwords; in development they include API keys, access tokens, and certificates.
Precise definitions vary by standard, but in AI and application development writing, the word usually covers the secrets needed to connect to an outside service.
Kinds of credentials
- Username and password
- API keys and secret keys
- Access tokens issued through OAuth
- SSH keys and private keys
- Certificates and passkeys
Lifetime and scope differ. Some API keys remain valid for a long time; access tokens often expire within minutes.
Authentication and authorization
Authentication establishes who is acting. Authorization decides what they may do. Holding valid credentials does not by itself mean permission to read or change everything.
The same split applies when delegating work to an AI agent. Supplying credentials and granting only the permissions needed are separate decisions: being able to sign in is not the same as being allowed to update or delete.
Why agents make this urgent
A chat assistant that only returns text needs no external credentials. An agent that operates mail, calendars, databases, GitHub, or a production server needs credentials for each of them.
Writing credentials into a prompt or into source code risks leaving them in conversation history, logs, shared files, and Git history. Using a secrets manager such as 1Password and supplying the value only at run time is the safer pattern.
Handling them safely
- Do not share them in plain text; use encrypted storage
- Grant the minimum permissions the job requires
- Use separate credentials per purpose and environment
- Review usage and revoke what is no longer needed
- Reissue immediately if a leak is possible
The useful question is not whether to give an AI credentials, but whether you can let it use them for a specific task without exposing the values.