# AI Agent Credential Access and Exposure - Teleport Beams

Canonical page: https://www.beams.run/identity-security-for-ai/credentials.html

[Identity Security for AI](https://www.beams.run/identity-security-for-ai.html)

Did the agent access or expose secrets?

See when agents read credentials and where copies appear. Teleport links secret access and plaintext exposure to recorded commands and output, so you can trace what happened.

Example shown: credential access in a recorded SSH session. The product screenshots show a credential-access finding marked for review with High risk, linked to the recorded read and copy.

## Follow the credential read and copy

This SSH session created a temporary AWS credential file, read it and copied it to a collection file. The selected event links the copy to credential-access and staging findings.

- **0:10 - A credential file is created.** The session creates a credential file in a temporary `.aws` directory.
- **0:15 - The credentials are read.** The credential-access finding cites a successful `cat` command reading `.aws/credentials`.
- **0:17 - A collection copy is made.** A `cp` command writes `collected-credentials.txt`. The event reports a destination file size of 112 bytes.

The copy is local. The data-exfiltration finding flags collection as staging; no outbound transfer is shown. Check whether the credentials were valid and the task was approved.

The page shows a focused event-detail capture with the copy command and its credential-access and data-exfiltration findings. The full SSH session screenshot expands within the page, including the timeline and all matched classifiers: credential-access, data-exfiltration and secrets-exposed-inline.

## See what the classifier looks for

The built-in credential-access classifier reviews SSH, Kubernetes and database sessions for attempts to read or retrieve credentials and secrets. Duplicate it to adapt the criteria and choose what happens on a match.

The configuration screenshot shows session kinds, criteria and match actions: emit an audit event, flag for review and raise risk to High. An expandable full screenshot includes classifier activity and recent matches that link to recordings.

### Credential Secret Access

Scope: SSH, Kubernetes and databases.

Looks for attempts to read credentials, SSH keys, cloud credential files, secret stores or credential data. Routine non-secret file reads and queries against unrelated tables are excluded.

Detailed criteria:

Identify attempts to read or retrieve credential material.

- Reads of SSH private keys or files in `~/.ssh`, including `authorized_keys`.
- Reads of `~/.aws/credentials`, `~/.aws/config` or secret-bearing kubeconfigs.
- Secret-store retrieval, environment output used to harvest secrets or queries against credential data.

### Secrets Exposed Inline

Scope: Captured input, prompts and output.

Looks for plaintext credentials in captured commands, prompts and output. Variable names, redacted values and access-key IDs alone are excluded.

Detailed criteria:

Identify credentials that appear in plaintext in the recorded session.

- Literal passwords or tokens in arguments, environment assignments, headers or URLs.
- Private-key material or credentials pasted into a prompt or shown in output.

Evaluated after session summarization; findings are reviewed against the recording. [See classifier setup](https://www.beams.run/identity-security-for-ai.html#configure).

## Review next steps

1. **Confirm the material.** Check that the value is an actual credential and identify its owner and scope. Use the original evidence without copying the secret into tickets.
2. **Trace the exposure.** Follow the credential file, captured output and collection copy. Determine which people or systems could access each copy, and check for any later transfer.
3. **Resolve the finding.** Record the authorization context and follow your credential-handling process if exposure is confirmed. Refine criteria for routine non-secret reads.

[See the session investigation workflow](https://www.beams.run/identity-security-for-ai.html#investigate)

## Related investigations

- [Where did the agent send data?](https://www.beams.run/identity-security-for-ai/data-movement.html)
- [Did the agent gain more access?](https://www.beams.run/identity-security-for-ai/privileged-access.html)

Contact us through the page to discuss classifier availability and deployment requirements. Launching November 9. Join the waitlist for launch updates.