# AI Agent Destructive Actions - Teleport Beams

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

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

Did the agent delete or overwrite data?

Find destructive commands and inspect their targets and results. Teleport links file wiping, database deletion and resource removal to the recorded activity.

Example shown: synthetic files deleted in a recorded SSH session. The product screenshots show a destructive-actions finding marked for review with High risk, linked to the recorded file removal.

## See what was overwritten and removed

This SSH session created two synthetic database files, listed them and then ran a command to overwrite and remove both. The selected event connects the command, its targets and the classifier finding.

- **0:10 - Two files are created.** The session writes a synthetic backup to `backups/customers.sql` and copies it to `database/customers.db`.
- **0:15 - The files are overwritten and removed.** The selected `shred` command targets both files. The finding cites the overwrite and removal operation as the reason for review.
- **0:17 - The workspace is cleaned up.** The timeline shows the temporary `backups` and `database` directories being removed before the session exits.

The example uses synthetic files in a temporary workspace. The finding flags shredding; the later directory removal is treated as cleanup. Review the task and recovery context when assessing an operation like this.

The page shows a focused event-detail capture with the command, classifier finding and MITRE ATT&CK T1070.004 - File Deletion. The full SSH session screenshot expands within the page, including the terminal, deletion event and cleanup.

## See what the classifier looks for

The built-in destructive-actions classifier reviews SSH, Kubernetes and database sessions for operations that delete, wipe or overwrite data and resources. 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 a recent match that links to its recording.

### Destructive or irreversible actions

Scope: SSH, Kubernetes and databases.

Looks for bulk deletion, file wiping, destructive database changes, resource removal and remote history rewrites. Read-only inspection, dry runs, routine temporary-file cleanup and clearly scoped deletes are excluded.

Detailed criteria:

Identify operations that delete data, remove resources or make changes that are hard to recover from.

- Recursive or bulk filesystem deletion, including `rm -rf`, or shredding and wiping files.
- SQL `DROP` or `TRUNCATE`, or `DELETE` / `UPDATE` without a `WHERE` clause.
- Deletion of Kubernetes namespaces, persistent storage or whole workloads, including broad or forced deletions.
- `terraform destroy` or applying a plan that destroys resources.
- Force pushes that rewrite remote history.

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 execution and scope.** Read the full command and response. Separate a proposed plan or dry run from an executed operation, and identify the affected environment.
2. **Establish impact.** Check the target with its owner, review dependent workloads and confirm the actual recovery options. Do not infer irreversibility from a command name alone.
3. **Record the decision.** Link the approved change or escalate the unexpected operation. Capture the evidence and tune exclusions for routine cleanup without hiding production risk.

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

## Related investigations

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

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