Back to evidence

Sanitized live incident

Confirmed compromise

Native source identity and targetable endpoints are private.

criticalconfirmed
Confidence
100%
First seen
Aug 23, 5:22:59 PM PDT
Evidence through
Aug 23, 7:44:59 PM PDT
AI status
Complete
True positive100% confidence

True positive. HTTP events [redacted] and [redacted] contained command-injection indicators and returned non-reflected uid=0/root process-identity output, proving root-level command execution in the responding workload even though the responses were HTTP 400. Process events [redacted] and [redacted] independently show a root dash shell followed by root id execution. Root processes also performed shared-resource mutation/access and sensitive-file-targeting activity. A processor-attributed outbound TCP flow to a private, remote-shell-class destination was observed, but it is not a proven request-to-socket or process-to-socket edge. The evidence does not establish host escape, persistence, lateral movement, command-and-control, or data theft.

Attack stage
Command-injection exploitation/RCE with post-exploitation discovery and resource access
Model
gpt-5.6-sol · 7 evidence calls

Observed impact

  • Root-level remote command execution and server identity disclosure in the responding workload (HTTP [redacted]; [redacted]).
  • Root shell and discovery-command execution observed in the correlated workload (process [redacted]; [redacted]).
  • Shared-resource mutation and access observed by root processes (process [redacted]; [redacted]).
  • Sensitive-file-targeting commands were observed in correlated workloads (process [redacted]; [redacted]).
  • An outbound TCP flow from the processor workload reached a private remote-shell-class destination (flow [redacted]).

Deterministic signals

Http.surface enumeration92%

Broad unauthenticated route and HTTP method enumeration observed

1800 observations · 10 http
Http.server command output100%

Exploit request received non-reflected process identity output

7 observations · 7 http
Process.correlated discovery command80%

A discovery process appeared in the correlated workload and request window

1970 observations · 12 process
Process.correlated shell spawn80%

A shell process appeared in the correlated workload and request window

1105 observations · 12 process
Process.correlated exit99%

A previously correlated process lifecycle exited

2014 observations · 12 process
Http.command injection attempt88%

Request contains shell metacharacters and command tokens

1069 observations · 12 http
Process.shared resource activity80%

A process modified an inventory-resolved resource attached to multiple workloads

9 observations · 9 process · 2 inventory
Process.correlated sensitive file command80%

A process command targeted a sensitive file in the correlated workload and request window

12 observations · 12 process
Process.correlated network client80%

An outbound-capable client process appeared in the correlated workload and request window

3 observations · 3 process
Flow.correlated new outbound destination70%

The correlated workload opened a flow to a destination not previously observed in the captured baseline

1 observations · 1 flow · 1 inventory

Explicit 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; the activity could have originated from an approved test, although the technical exploit and root execution are proven.
  • Configured routing and time-window correlation do not establish a unique HTTP-request-to-process parentage edge.
  • The process evidence establishes workload-level execution, not host escape or host compromise.
  • The shared resource's changed content and the effect on other attached workloads are not shown by the bounded evidence.
  • Sensitive-file command targeting does not establish what content was successfully read, retained, or exfiltrated.
  • Conntrack and inventory attribution do not establish which request or process opened the outbound socket; the private remote-shell-class flow therefore does not by itself prove lateral movement or command-and-control.
  • No cited evidence proves persistence, credential theft, data exfiltration, or the full compromise scope.

Recommended actions

  1. Immediately contain and replace the affected processor and image-host workloads from known-good images; preserve current runtime evidence and snapshots before destructive remediation where feasible.
  2. Disable or restrict the vulnerable API route and remediate the command-injection flaw; validate input handling and remove shell invocation from request processing.
  3. Rotate credentials, tokens, and secrets available to the affected workloads, especially values exposed through environment files or sensitive files.
  4. Inspect shared resource [redacted] for unauthorized changes, restore verified content, and assess every workload attached to it.
  5. Identify the private remote-shell-class destination through internal asset inventory, review both endpoints' logs, and block or isolate the path if it is not explicitly authorized.
  6. Review the complete process lineage and filesystem changes across both correlated workloads for additional payloads, persistence mechanisms, and accessed data.
  7. Reduce workload privilege: run as a non-root identity, use a read-only filesystem where possible, minimize mounted shared resources, and enforce least-privilege network egress.
  8. Preserve and correlate application, gateway, workload, identity, and destination-host logs for the full incident window before closing scope.