The polygraph has been used in UK national security vetting since 2013. It remains in use despite the 2003 National Academy of Sciences finding that it is "not useful for employee security screening" at the national security level, and despite published false positive rates of 10–30% that mean a material proportion of screened personnel are incorrectly flagged. The question for procurement is not whether the polygraph is imperfect — it is — but what the alternative measurement framework offers and whether it addresses the underlying problems.
The core difference: what each system actually measures
The polygraph does not measure deception. It measures the physiological correlates of anxiety — skin conductance (galvanic skin response), respiratory rate, blood pressure, and pulse. The implicit assumption is that deception produces measurable anxiety. This assumption has two fundamental problems: anxious innocent subjects produce the same physiological signals as deceptive subjects, and countermeasure-trained deceptive subjects can suppress these signals without suppressing the underlying deception.
EchoDepth measures FACS Action Units — the specific facial muscle movements that are the direct physical expression of emotional states. Rather than measuring a peripheral proxy, it measures the output of the emotional system at the face: 44 Action Units per frame, at up to 30fps, producing VAD dimensional scores (Valence, Arousal, Dominance) against each individual's established baseline. This is a different measurement layer, not a refinement of the same measurement.
Side-by-side comparison
| Capability | Polygraph | EchoDefence |
|---|---|---|
| What it measures | Peripheral physiological arousal (GSR, respiratory rate, blood pressure, pulse) | 44 FACS Action Units — direct facial muscle expression of emotional state |
| Output type | Deceptive / Non-Deceptive / Inconclusive label | VAD dimensional scores per frame — Valence, Arousal, Dominance — with AU evidence |
| False positive rate | 10–30% (NAS 2003; field conditions toward higher end) | Individual baseline deviation scoring reduces population-level false positive artefacts |
| Countermeasure susceptibility | High — relaxation and self-justification techniques suppress peripheral arousal | Lower — peripheral suppression does not suppress facial AU activations. Masking is detectable. |
| Output auditability | Examiner interpretation — no reproducible evidence trail | Per-frame JSON: AU evidence, VAD coordinates, confidence weights, baseline deviation |
| Contact with subject | Requires physical electrode attachment | Camera only — no physical contact |
| Continuous monitoring | Point-in-time interview only | Continuous — can monitor full interview duration at frame level |
| Cultural calibration | Technique calibrated on Western populations | 14-cohort training data, 6 countries |
| Suppression detection | No — suppression defeats peripheral signal | Yes — micro-expression trajectory analysis detects AU suppression patterns |
| UK data residency | Varies by examiner and organisation | Yes — default configuration, all processing on-premise |
| SCIF / air-gap deployable | Yes (analogue equipment) | Yes — zero outbound network connections required |
| UK-developed | No — US origins, US vendors dominant | Yes — Cavefish, Cardiff, Wales |
The accuracy problem: both error types
The polygraph's accuracy problem is not one-sided. Both error types — false positives and false negatives — are significant and well-documented.
A false positive rate of 15% applied to a vetting population of 1,000 personnel produces 150 incorrect deceptive findings against innocent subjects. Each represents a career interrupted, a clearance challenged, and an HR and legal process initiated on the basis of a physiological stress response the polygraph cannot distinguish from deception. For individuals from backgrounds or cultures where authority-based interrogation carries particular associations, the false positive rate is likely higher.
A false negative rate of 10–20% in laboratory conditions — demonstrably higher in real-world conditions where subjects have access to published countermeasure literature — means that a material proportion of deceptive subjects pass the screening. The Ames and Hanssen cases are the most extensively documented examples, but they illustrate a systematic rather than exceptional vulnerability.
EchoDepth's individual baseline deviation scoring does not eliminate false positives — no system does — but it addresses the mechanism that generates the highest-rate polygraph false positives: population-level threshold exceedance from anxiety in innocent subjects. By comparing each individual's emotional state pattern against their own established baseline, rather than against a population norm, the system reduces the contribution of individual anxiety variation to the false positive rate.
"Almost a century of research in scientific psychology and physiology provides little basis for the expectation that a polygraph test could have extremely high accuracy."
— National Academy of Sciences, The Polygraph and Lie Detection (2003)The countermeasure problem
Polygraph countermeasures are published, accessible, and effective. The technique works against the CQT (Control Question Technique) by reducing the differential arousal signal between relevant and control questions — either by suppressing arousal at relevant questions or augmenting arousal at control questions. Physical techniques include biting the tongue, pressing the toes, and controlled breathing. Cognitive techniques include relaxation, self-justification narratives, and reframing. A determined adversary with advance preparation time has access to all of these.
EchoDepth's countermeasure surface is different. Controlling peripheral nervous system arousal — the mechanism that defeats the polygraph — does not suppress the facial Action Unit activations that accompany emotional states. These activations occur in the facial musculature through a pathway partially independent of peripheral arousal suppression. Deliberate facial masking — consciously controlling facial expression — is possible, but it is itself detectable: the effort to hold a neutral expression produces AU patterns (specifically AU1, AU4, AU7 activation without natural expression flow) that deviate from the individual's baseline authentic expression pattern.
This is not a claim of countermeasure-immunity. No assessment tool is. It is a claim that EchoDepth's countermeasure surface is distinct from the polygraph's, and that the techniques widely published for defeating the polygraph do not straightforwardly transfer.
Auditability and UK procurement requirements
UK government AI procurement increasingly requires explainability, auditability, and challenge mechanisms. A polygraph examiner's finding — "deceptive" — is a professional interpretation that cannot be decomposed or reviewed. The evidence trail ends at the examiner.
EchoDepth's per-frame structured output provides a complete evidence chain:
- Timestamp (ISO 8601)
- Active Action Units with intensity scores
- VAD coordinates: V: −0.52, A: +0.81, D: +0.14
- Confidence weight (0–1)
- Baseline deviation magnitude for each dimension
- Question or event tag (linked to interview timeline)
Any finding — elevated arousal at a specific question, transient VAD excursion, suppression pattern — can be reviewed, contextualised, and challenged with reference to this structured log. This satisfies the evidentiary requirements that a discrete label output cannot.
Deployment: complement or alternative
EchoDefence can be deployed in two configurations in UK defence personnel security contexts. As a complement to existing interview-based vetting, it provides a continuous emotional state layer during assessments that currently rely on unstructured human observation — adding AU-level evidence to the existing process without replacing it. As a standalone alternative for specific screening populations or assessment types, it provides a fully evidenced, auditable output that addresses the documented accuracy and countermeasure limitations of polygraph screening.
For UK national security vetting contexts, any alternative to polygraph screening requires evaluation within the existing NSV framework. EchoDepth's procurement pathway documentation covers the evaluation and approval route for UK government deployment. For a technical briefing, contact the EchoDepth Defence team.
For further accuracy detail, see polygraph false positive and false negative rates and the scientific evidence on polygraph false negatives.
FACS-grounded credibility assessment for UK defence vetting
44 Action Units. VAD dimensional output. Per-frame audit trail. SCIF-deployable. UK-developed and supported.