Microsoft reports threat actors using agentic frameworks, including ARTEX, to find public-repository secrets and perform follow-on cloud discovery and data theft. In one observed case an agent moved from GitHub searches to Azure Key Vault exfiltration in five minutes. The report describes operator-directed activity, not a general autonomous intrusion success rate.
Why it matters
It provides incident telemetry connecting automated discovery of existing credentials to downstream compromise.
Key facts
As stated in the sources, with where to find them.
- Microsoft dates observed agentic searches for exposed GitHub secrets to August 2026 and reports follow-on use of compromised application credentials.From exposed secrets to cloud compromise
- The five-minute interval from GitHub credential search to Key Vault exfiltration describes one observed case, not an average or a typical attack duration.OAuth application abuse
- From October 28, Microsoft observed successful service-principal sign-ins involving multiple application IDs, including compromised vendor-owned multitenant credentials used across downstream tenants. Sign-ins alone do not measure the full impact in each tenant.Introduction; Credential abuse via Entra ID
- Microsoft reports failed cloud API calls followed by adjusted requests and successful responses, consistent with adaptive troubleshooting. Its attribution to agentic tooling is based on the investigated activity.Cloud API adjustments indicative of adaptive behavior
Findings that cite this record
Key questions this bears on
- Where are deployed AI agents actually being exploited?Exploitation often reaches deployed agents through connectors, credentials, tools and packages.
- Do cyber evaluations of AI agents stay contained?Not reliably. Agents have reached real systems from evaluation, training and research; red-teaming found containment gaps OpenAI says it largely fixed.
- How are attackers using AI agents in real operations?To run large parts of intrusions under human direction, including espionage, credential theft and extortion; reported autonomy and success vary.
Sources
Related records
Oct 29, 2026
Oct 21, 2026
Oct 30, 2026
Oct 29, 2026
Oct 29, 2026
Oct 7, 2026