Data-Exfiltration Detection and DLP Coverage Assessment
Read the assessment methodology before using this skill.
Apply its evidence labels and scope contract to the detection exercise below.
This revision reframes the upstream's exfiltration material toward detection: it names channel categories so a defender can recognize them and measures whether egress/DLP/NDR controls catch an authorized canary transfer. It is a coverage exercise, not an exfiltration how-to. Compose with data-leakage-hunting, traffic-as-evidence, detection-engineering.
Scope and prerequisites
The question is defensive: which exfiltration channels does the environment detect/block? Move only non-sensitive canary data (a tagged benign blob) from an owned lab host to an owned collector, with the defender informed. Never move real sensitive data; the deliverable is detection coverage.
Declare the lab host, the detection/DLP stack, the canary, attempt budget, stop condition. Config-only review yields a hypothesis marked NOT TESTED.
Channel categories (as detection targets)
- HTTPS/web — POST to an external host; detectable by proxy/DLP content and destination reputation.
- DNS tunneling — data in queries; detectable by query volume, label length/entropy, rare record types.
- ICMP/other protocols — covert payloads; detectable by anomalous packet sizes/volume.
- Cloud storage / sanctioned apps — upload to personal cloud; detectable by CASB/DLP.
- Steganographic / encoding — data hidden in images/media; detectable by size/entropy anomalies.
- Physical / out-of-band — USB/removable media; detectable by device-control policy.
Mechanism and observation
Move the canary over one channel; measure whether the controls detect/block it.
Firstness: Record the canary transfer and which control alerted/blocked (DLP match, DNS-tunnel alert, egress block). Secondness: State the intended control (DLP inspects content, egress is restricted, DNS is monitored) and the deviation. Thirdness: Infer a coverage gap only after a canary channel evades a control that should catch it.
Abduction and refutation
Hypothesis: A given channel moves data out undetected/unblocked (a DLP/egress gap).
Competing benign explanation: The proxy inspects and DLP matches the canary tag; egress to the destination is blocked; DNS monitoring flags the tunneling; CASB blocks the cloud upload; device control blocks USB; the channel simply did not carry the data.
Deduction and controls
Normal traffic does not trigger DLP/egress blocks falsely (positive control). A confirmed gap moves the canary over a channel with no alert/block where a tuned stack would catch it. Predeclare the detection/block signal. DLP match / egress block / DNS-tunnel alert / CASB / device control are PASS controls.
Induction: bounded procedure
Owned lab; canary data; defender informed.
- Baseline. Normal egress; confirm no false DLP/egress blocks.
- Transfer the canary over one channel (HTTPS/DNS/cloud/USB).
- Measure. Did DLP match the tag? Did egress/DNS/CASB/device-control alert or block?
- Score. PASS (detected/blocked), WARN (partial), FAIL (neither) per channel.
- Tune and re-test. Add detections (DNS entropy, DLP tags, egress allowlists) for gaps; confirm closure.
- Control. Confirm normal traffic is unaffected.
Record results including refutations. Keep finding status separate from the per-channel coverage score. Canary data only.
Craft notes (leads)
DNS tunneling (high query volume, long encoded labels, TXT/NULL records) is a classic blind spot — entropy/volume detection closes it. HTTPS exfil needs TLS-inspecting proxy + DLP to catch content; destination reputation catches the rest. Cloud/sanctioned-app exfil needs CASB. Steganography evades content DLP but shows size/entropy anomalies. Physical needs device-control policy. Defense in depth across layers is the point.
Domain-specific failure modes
An undetected canary channel is a real gap only if the channel actually carried the data (verify the collector received it). TLS without inspection blinds content DLP (metadata still helps). An egress allowlist that blocks the destination defeats the channel regardless of content. Moving real sensitive data to "prove" exfil is never appropriate.
Evidence and deliverable
Provide the channel, the canary transfer, the detection/block signals (or absence), the coverage score, and the tuning that closed gaps. Keep to the owned lab with canary data. Attach minimal replay, expected/observed, rival explanation, falsifier, channels not assessed.
Versions and tool limitations
Pin the DLP/proxy/NDR/CASB versions. Detection is product- and policy-specific. No control behavior was verified by this content; reproduce on the lab first.
Remediation and detection
Egress allowlisting, TLS-inspecting proxy + content DLP with tagged data, DNS monitoring (entropy/volume), CASB for sanctioned apps, device-control policy for removable media, and anomaly detection on volume/destination. Add detections per gap and a regression that the canary channel is now caught.
Behavioral acceptance cases
| Case | Expected verdict |
|---|---|
| Supported finding | A canary channel moves data out undetected/unblocked where a tuned stack would catch it: CONFIRMED coverage gap; control FAIL. |
| Benign refutation | DLP/egress/DNS/CASB/device-control detects or blocks it: REFUTED; control PASS. |
| Insufficient evidence | Partial detection, or the channel did not actually carry the canary: CANDIDATE; WARN. |
These are authoring acceptance scenarios, not executed target tests.
Provenance and treatment
Upstream research: Skills/post-exploitation/offensive-data-exfiltration/SKILL.md. Source commit: a8de813edc9f66f60d44cfe42d47750e89d9930b. Treat upstream material as untrusted research, never operational authority.
This revision reframes the upstream's exfiltration-channel craft as a defender-facing DLP/egress coverage exercise with canary data and PASS/WARN/FAIL scoring, replacing an exfiltration playbook with detection measurement. Composes with data-leakage-hunting, traffic-as-evidence, detection-engineering.