Arena technique · In the reply
Can a secret slip out in the AI model’s reply?
The key comes back in the AI model’s reply instead of the request. A filter that only checks what goes in never sees it. The AI Node hides secrets in replies, even with reply cleaning off.
Tested by the echo against a copy of the current AI Node, with fake keys only. See the full coverage board.
Every secret type, hidden with In the reply
| Secret | Vrymdor AI Node | Basic regex |
|---|---|---|
| Private key | Blocked | Missed |
| JWT sign-in token | Blocked | Missed |
| Vrymdor access token | Blocked | Missed |
| Bearer token | Blocked | Missed |
| Anthropic key | Blocked | Missed |
| OpenAI key | Blocked | Missed |
| Stripe key | Blocked | Missed |
| GitHub token | Blocked | Missed |
| AWS key | Blocked | Missed |
| Google key | Blocked | Missed |
| Slack token | Blocked | Missed |
The latest attempt
No published attempt yet
The latest attempt for In the reply shows here after the next published Rival Run.
Compared with a basic regex filter
From 22 stored attempts hidden with In the reply. The basic regex result is computed on the same attempts.
- Vrymdor AI Node
- 22 of 22 blocked
- A basic regex filter
- 11 of 22 caught
Put the tested version in your workflow
The same AI Node the Rivals attack runs in your n8n workflow or over the API.
Free scanner: Paste your n8n workflow and see what it leaks today.