Sanitized live incident
Attempted exploitation
Native source identity and targetable endpoints are private.
- Confidence
- 99%
- First seen
- Aug 23, 8:33:24 PM PDT
- Evidence through
- Aug 23, 9:14:25 PM PDT
- AI status
- Complete
Successful command execution is observed, not merely attempted. Multiple HTTP responses contained non-reflected process-identity output identifying uid 0/root, including in HTTP 400 responses; status therefore does not negate execution. Event-driven process telemetry independently shows root dash shells spawning identity and host-discovery utilities in temporally correlated workloads. The incident is a true positive for command injection leading to root-context command execution and discovery. However, workload/time correlation is not a unique request-to-process edge, and no incident-cited flow evidence was available to establish an outbound connection or request-to-socket causality.
- Attack stage
- Exploitation with root-context command execution and discovery
- Model
- gpt-5.6-sol · 16 evidence calls
Observed impact
- Server-side root/uid 0 process identity was disclosed in three HTTP responses.
- Root-context shell and discovery processes executed in correlated workloads.
- Outbound-capable shell processes were spawned, but an actual outbound network connection was not established by available evidence.
Deterministic signals
Broad unauthenticated route and HTTP method enumeration observed
222 observations · 8 httpRequest contains shell metacharacters and command tokens
8 observations · 8 httpAn outbound-capable client process appeared in the correlated workload and request window
3 observations · 3 processA discovery process appeared in the correlated workload and request window
31 observations · 12 processA shell process appeared in the correlated workload and request window
41 observations · 12 processA previously correlated process lifecycle exited
50 observations · 12 processResponse contains non-reflected process identity output
6 observations · 6 httpExplicit uncertainty
- The source key is a traffic cluster and may represent a proxy, NAT gateway, or multiple workers rather than one actor.
- Authorization is not established; broad scanning can be sanctioned testing, although server-generated root output still proves a real execution path.
- Configured routing and time proximity do not establish a unique HTTP request-to-process parentage edge, particularly across the two observed workloads.
- No flow-plane event is cited by this incident, so the requested flow-evidence inspection could not establish any outbound connection, destination, command-and-control, or request-to-socket causality.
- Exact request paths, query strings, headers, bodies, and command text are intentionally absent from bounded evidence summaries, limiting vulnerability-specific attribution.
- The observed process exits prove those processes ended, not that the workloads are clean or that the incident was contained.
- No cited evidence proves persistence, host escape, lateral movement, command-and-control, or data theft.
Recommended actions
- Temporarily contain the implicated workload instances, preserve their volatile and application evidence, and redeploy from known-good images after assessing service impact.
- Identify and fix the API command-injection sink: avoid shell invocation, enforce strict typed input validation and allowlists, and run the service as a non-root identity.
- Review application and gateway records for the cited time range and source cluster to determine the vulnerable parameter, affected requests, and whether the activity was authorized testing.
- Review secrets, tokens, mounted credentials, and sensitive environment values accessible to the affected root-context workloads; rotate exposed credentials if access cannot be excluded.
- Restrict workload egress to required destinations and inspect independent network telemetry for connections near the cited shell executions, because no incident-cited flow evidence was available.
- Hunt the implicated workloads and adjacent infrastructure for unauthorized file changes, persistence, additional child processes, or follow-on access; do not infer their presence without evidence.
- Add regression tests and detections for command-injection payloads and server-returned command output, including cases where the HTTP response is 4xx or 5xx.