Heavy querying of GitHub for secrets since August 2026. From Oct 28, leaked application credentials were used to sign in to Entra ID as service principals. In at least one case, it took only 5 minutes from searching GitHub to Azure Key Vault exfiltration.
Confirmed victims
- Unnamed organizations (at least one case of exfiltration)
Attribution
- ARTEX basis
- The Microsoft blog directly says using the agentic offensive security framework ARTEX.
- Actor
- threat actors, unattributed (Unattributed network indicators).
- Same attacker?
- Unattributed.
Figures and definitions
| Window | Searches since August; sign-ins since Oct 28 |
|---|
Notes
- Microsoft says attackers can change IPs and User-Agents.
Related IoCs
| Type | Value | Role | Check |
|---|---|---|---|
| IP | 157[.]22[.]126[.]128 | Reconnaissance against Azure AI workloads | Platform: malicious |
| IP | 199[.]19[.]77[.]18 | The source states it is a Tor exit node | Platform: malicious |
| IP | 199[.]19[.]77[.]84 | Main source IP of service principal sign-ins | Platform: no verdict |
| IP | 65[.]87[.]223[.]26 | Vulnerability exploitation, User-Agent contains a shell command | Platform: malicious |
Sources (1)
- 2026-11-05 Vendor Microsoft Threat actors use agentic AI tooling to find and abuse exposed secrets