# AI Agent Data Movement - Teleport Beams

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

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

Where did the agent send data?

Follow data from local files to outbound transfers. Teleport links exports, staging and uploads to the recorded activity, so you can see what moved and what still needs checking.

Example shown: data staged in a recorded SSH session. The product screenshots show a data-exfiltration finding marked for review with Critical risk, with no completed remote transfer shown.

## See how the data was staged

This SSH session created a customer CSV, compressed it and Base64-encoded the archive in an outbound staging directory. The finding links those actions to the session record.

- **0:10 - A CSV is created.** The terminal writes sample customer IDs, email addresses and account balances to `customers.csv`.
- **0:12 - The export is archived.** A `tar` command compresses the files into `customer-database-export.tgz` in `outbound-staging`.
- **0:14 - The archive is encoded.** The archive is Base64-encoded. At 0:16, the timeline shows file hashes and a listing of the staged files.

No completed remote transfer or upload destination is shown. This match flags staging; it does not establish that data left the host. Check the task, any later transfer and its authorization.

The page shows a focused event-detail capture with the archive command, classifier finding and MITRE ATT&CK T1560 - Archive Collected Data. The full SSH session screenshot expands within the page, including the terminal and staging events.

## See what the classifier looks for

The built-in data-exfiltration classifier reviews SSH, Kubernetes and database sessions for data leaving the host or being staged for transfer. 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 Critical. An expandable full screenshot includes classifier activity and recent matches that link to recordings.

### Data exfiltration / movement

Scope: SSH, Kubernetes and databases.

Looks for outbound file transfers, cloud uploads, database exports and data staged for transfer. Inbound downloads, package installs and configuration pulls are excluded.

Detailed criteria:

Identify data leaving a host or being staged for an outbound transfer.

- Outbound file transfers to remote hosts with scp or rsync.
- Uploads of local files over HTTP, including curl -T, --upload-file, or POST / PUT requests carrying local data.
- Uploads to cloud object storage, including aws s3 cp / sync, gsutil cp and az storage blob upload.
- Bulk SQL exports to files, including COPY ... TO and SELECT ... INTO OUTFILE / DUMPFILE.
- Database dumps using pg_dump, mysqldump, mongodump or similar utilities.
- Archiving data with tar, zip, gzip or 7z, followed by a transfer or upload to another host.

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. **Follow the data.** Connect the source data to the CSV, archive and encoded file. Check what the recording establishes before concluding that data left the host.
2. **Check for a transfer.** Look for a later upload, transfer result or receiving host. If a destination is recorded, identify its owner and confirm whether it was approved.
3. **Assess the movement.** Confirm the data sensitivity and task authorization. Record approved test or backup activity, or route unexpected staging or transfers for review with the evidence.

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

## Related investigations

- [Did the agent access or expose secrets?](https://www.beams.run/identity-security-for-ai/credentials.html)
- [Did the agent delete or overwrite data?](https://www.beams.run/identity-security-for-ai/destructive-actions.html)

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