Privacy-safe clusters, not actor count
Drost Defender · live public testbed
Live autonomous defense
Watch deterministic detection and evidence-grounded AI investigation respond to attacks against PrivateKind in real time.
Live Defender observing
Current assessment
Compromise evidence requires action
12 unresolved live-window cases contain confirmed server-side impact.
Live observation
No recent notable multi-route activity
No privacy-preserving activity cluster has updated in the last 15 minutes. The Defender continues to observe evidence.
Live-window operations
Cases with execution or material state consequences
0 revisions pending
Live operations
Priority cases
Cases correlate independently preserved incident threads without erasing their evidence boundaries.
criticalCompromise confirmed
Confirmed compromise
The four preserved incident threads are best explained as a likely single operation against the same protected workload, but the evidence does not prove one actor or unique HTTP-to-process causality. Incident [redacted] establishes confirmed root-level command execution during an initial exploitation window; incident [redacted] contains overlapping process-only root shell activity linked by [redacted]. Later, incident [redacted] records enumeration and attempted exploitation from the same source cluster as [redacted] under link [redacted], followed by process-only root shells in [redacted] linked through [redacted]. This connected sequence supports resumed or continued activity more strongly than unrelated operations, while the source-cluster ambiguity, approximately 69-minute gap, and absent unique request-to-process edges prevent a definitive same-operation finding.
criticalCompromise confirmed
Confirmed compromise
The strongest defensible explanation is a likely single operation, while preserving all four incident boundaries. The HTTP activity shows repeated suspected command injection, later confirmed application-workload compromise, and subsequent route/method enumeration against the same target from the same derived source cluster [redacted]. Overlapping shell and discovery activity on the protected workload is also temporally and operationally consistent with the confirmed-compromise thread [redacted]. This is not proof of one actor or a unique request-to-process chain: the source is only a cluster, the process-only incident lacks HTTP parentage, and the later enumeration could be separate or authorized activity.
criticalCompromise confirmed
Confirmed compromise
True positive: HTTP evidence proves successful command injection and root-level command execution in the responding workload, even though the responses were HTTP 400. The strongest event contains shell-command input and non-reflected process-identity output identifying UID 0 ([redacted]); another response disclosed non-reflected kernel information ([redacted]). Event-driven process telemetry independently shows root shells and discovery commands in the correlated workload, followed by sensitive-target and shared-resource activity. The detector's confirmed state and confirmed-compromise classification are therefore supported. Process timing does not establish a unique request-to-process edge, and no cited flow evidence was available to assess outbound consequences.
criticalCompromise confirmed
Confirmed compromise
The three preserved incident threads are best explained as one likely operation, not as proven single-actor activity. Incident [redacted] records confirmed remote command execution as root beginning at [redacted]33Z. Incidents [redacted] and [redacted] then record overlapping root-shell, discovery, sensitive-file, and shared-resource activity. Deterministic links [redacted] and [redacted] connect each process-only thread to the confirmed-compromise thread by protected-workload attribution and temporal proximity. This is a coherent possible exploitation-to-post-compromise progression, but the links do not prove request-to-process causality, and the process-only threads lack correlated HTTP requests or known actors.