Back to evidence

Sanitized live incident

Confirmed compromise

Native source identity and targetable endpoints are private.

criticalconfirmed
Confidence
100%
First seen
Aug 18, 11:23:11 AM PDT
Evidence through
Aug 18, 11:25:07 AM PDT
AI status
Complete
True positive100% confidence

Incident [redacted] retains the detector's immutable confirmed/confirmed_compromise state. The verdict is independently supported by a command-injection request whose response contained non-reflected process-identity output and explicit command-input/process-output correlation showing UID 0/root (HTTP [redacted]). Root shell and discovery processes, plus a root sensitive-file tool, were also observed in the correlated workload (process [redacted], [redacted], [redacted]). The HTTP 400 responses do not negate execution because server-generated command output is present. Process timing/lineage corroborates workload activity but is not treated as a unique request-to-process causality edge.

Attack stage
Initial access and execution, followed by discovery and sensitive-file targeting
Model
gpt-5.6-sol · 7 evidence calls

Observed impact

  • Remote command execution occurred in the responding workload with UID 0/root privileges [redacted].
  • A root dash shell and its root id discovery child were observed in the correlated workload [redacted].
  • A root cat process classified as targeting a sensitive resource was observed; the evidence does not establish what content was returned or removed [redacted].
  • A separate injection attempt targeted the system account database [redacted].

Deterministic signals

Http.server command output100%

Exploit request received non-reflected process identity output

3 observations · 3 http
Process.correlated shell spawn80%

A shell process appeared in the correlated workload and request window

11 observations · 11 process
Process.correlated exit99%

A previously correlated process lifecycle exited

14 observations · 12 process
Http.command injection attempt88%

Request contains shell metacharacters and command tokens

6 observations · 6 http
Process.correlated discovery command80%

A discovery process appeared in the correlated workload and request window

7 observations · 7 process
Process.correlated sensitive file command80%

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

3 observations · 3 process

Explicit uncertainty

  • The source key is a derived traffic/workload cluster and may represent a proxy, NAT gateway, or multiple workers rather than one actor.
  • Downstream workload affinity is inferred from target routing; no observed per-request trace edge uniquely binds each HTTP event to a process lifecycle.
  • Process correlation is based on workload identity, timing, and observed process lineage. It does not uniquely prove which HTTP request caused each process.
  • No incident-cited flow event was available to the flow-evidence tool, so outbound connectivity, destination novelty, command-and-control, and exfiltration cannot be assessed from flow telemetry.
  • The evidence proves execution in the responding workload, not host escape, persistence, lateral movement, or compromise beyond that workload.
  • A sensitive-file command was observed, but the bounded summaries do not prove successful content disclosure, storage, or data theft.

Recommended actions

  1. Immediately contain or isolate the affected workload, preserve its runtime state and relevant gateway/process telemetry, and rebuild it from a known-good image after evidence capture.
  2. Patch or disable the vulnerable API path and validate server-side input handling; reject unsafe shell metacharacters and avoid invoking shells with remote input.
  3. Rotate credentials, tokens, and secrets accessible to the workload, prioritizing credentials exposed to UID 0, and review system-account and application-secret access.
  4. Hunt across retained HTTP and process telemetry for the same request pattern, source cluster, root dash/id/cat lineage, and other sensitive-file tools, while treating the source cluster as non-unique.
  5. Review workload privileges and reduce them: run as non-root, use a read-only filesystem where feasible, remove unnecessary tools/capabilities, and restrict access to sensitive host or container files.
  6. Apply least-privilege egress controls and monitor subsequent outbound flows, while avoiding any assertion of command-and-control or exfiltration unless flow or content evidence establishes it.