Implementation guide
OWASP AISVS v1.01
Monitoring, Logging & Anomaly Detection · Detection and Alerting
Watch AI traffic for covert channels
Verify that LLM API traffic is monitored for covert-channel indicators and communication signatures to identify malware and command-and-control (C2) activity.
OWASP AISVS v1.01, requirement 12.2.6
In plain English
Malware has begun using model APIs as a hidden channel: fetching instructions, generating code on demand, or sending data out in traffic that looks like ordinary AI use. Watch traffic to model APIs for signs of this. This is a Level 3 control.
What good looks like
- Traffic from your network to model API providers is monitored.
- The approved clients, accounts and API keys are known, and anything else stands out.
- Indicators raise alerts: unknown processes or hosts calling model APIs, unapproved keys, encoded payloads, or regular beacon-like timing.
- Alerts are investigated by the security team.
How teams usually do it
- Egress monitoring and proxy rules for model API domains, with access allowed only through your gateway.
- Endpoint detection rules for processes that call model APIs unexpectedly.
- Inspection of prompts at the gateway for encoded blobs and command-like content.
Suggested evidence
- Monitoring rules.
- The inventory of approved clients.
- Sample alerts.
Common mistakes
- Treating all traffic to model APIs as normal business use.
- Allowing direct access to provider APIs from any machine, so there is no baseline to compare with.
Not sure which requirements apply to your system? Take the assessment, or browse every guide in this chapter.
Requirement text from the OWASP AI Security Verification Standard, licensed under CC BY-SA 4.0. This guidance is shared under the same licence.