A retrospective review of "Can the police monitor a suspect's brain during interrogation? A new report issues a stark warning" Allan McCay and David Hamer (University of Sydney), The Conversation, 17 November 2026 https://theconversation.com/can-the-police-monitor-a-suspects-brain-during-interrogation-a-new-report-issues-a-stark-warning-295555
This review is written as if read in 2066, on one assumption: technology that can read from, and act upon, the human central nervous system works reliably, is widely deployed, and is a mainstream legal and social concern.
The scope is deliberately the whole central nervous system β brain, spinal cord, and retina and optic nerve β not just the brain, and covers technologies that write (alter neural activity) as well as those that read it.
The question is not "does it work?" but "if it does, how well did the 2026 warning prepare us?"
Terms used
| Term | Meaning |
|---|---|
| EEG | Electroencephalography β recording the brain's electrical activity through sensors on the scalp |
| P300 | A brain-wave response appearing roughly 300 milliseconds after a person sees something they recognise or find significant |
| CIT | Concealed Information Test β shows crime-related items mixed among irrelevant ones and looks for a recognition response (also called the Guilty Knowledge Test) |
| BEOS | Brain Electrical Oscillation Signature profiling β a separate Indian technique claiming to detect "experiential knowledge" |
| Probe | The crime-related item a suspect is shown during a recognition test |
| CNS | Central nervous system β the brain, spinal cord, and retina and optic nerve |
| Terahertz (THz) waves | Electromagnetic radiation between microwaves and infrared light (0.1β10 THz); non-ionising, and able to interact with proteins in nerve cells |
| Neuromodulation | Changing nerve activity from outside β "writing" to the nervous system rather than "reading" it |
The article's strongest and most prescient section. It identified that neural evidence would reach Australian courts not by statute but by judicial discretion β through an evidence regime that favours admission, and a judge's power to admit improperly obtained evidence where no violence was used. That is precisely how a new evidence type arrives before Parliament catches up.
It noticed that the Lie Detectors Act was written for polygraphs and would not reach image-recognition testing. Recognition testing was always the more workable technique β and it sat outside the only law that might have stopped it.
The report's list β no compulsory or coercive neural lie detection, memory retrieval, or guilty-knowledge testing; no neurotechnology for interrogation, behavioural compliance or involuntary treatment β remains a sound minimum.
The article's load-bearing claim is that the technology is unproven, and therefore police use is premature. An argument built on "not yet" expires the moment the technology arrives. Read from 2066, the central thesis is simply obsolete; only the legal and rights sections survive.
Lesson: a safeguard justified by "it doesn't work yet" stops protecting anyone the day it works.
The article pictures police placing a headset on a suspect in an interview room. In 2066 the neural data already exists β on consumer headsets, earbuds and wellness wearables. Police don't fit a device; they request what yours already recorded. The UN Secretary-General's October 2026 report saw this coming, warning that signals from a gaming headset or wellness gadget could become courtroom evidence. The article's consent-versus-coercion framing has no answer to data a person handed over voluntarily, for other purposes, years earlier.
While calling the technology unproven, the article also describes brain-computer interfaces as able to "transfer a person's thoughts", and repeats the report's term "memory retrieval". Neither describes recognition testing, which detects whether a shown item is familiar. Overstating the threat is the mirror image of vendor overclaiming β and equally unhelpful to courts.
Its main case study (Aditi Sharma) involved BEOS, not P300 recognition testing. BEOS claims to detect participation in an event β a far stronger claim than P300 testing makes β and was criticised by an Indian expert committee. Using it as the example blurs a modest, well-studied technique with an inflated, poorly validated one.
The most corrosive long-term effect. Once a test is reliable, declining it looks like guilt β as refusing a breath test does. The right to silence can survive in law while losing its value in practice: juries, employers and the public draw the inference regardless. The article treats compulsion as the threat; in a world where the test works, social pressure does the compelling.
Accused people would demand testing to clear themselves. This splits defendants into those who can be cleared and those who can't or won't be tested β and hands prosecutors a powerful line: "The defendant was offered a test and declined."
A recognition device registers every strong response, not just the crime-related ones. Research already shows people with substance use disorders produce larger P300 responses to drug-related cues. A test for knowledge of a murder can therefore expose a person's drug history β self-incrimination by accident, on matters nobody asked about. This falls hardest on people who use drugs, who are heavily over-represented in police interviews.
The article deals with past crimes. The real 2066 battleground is neural risk scoring β in bail, parole and sentencing. The UN Secretary-General had already flagged the use of brain data to predict future offending.
Verifying a witness's memory sounds benign until a traumatised victim is pressured to "prove" they aren't lying. The article considers suspects only.
The article and the report deal almost entirely with extracting information. The research frontier already included technologies that alter neural activity without implants or drugs. Terahertz waves, for example, sit in the energy range of hydrogen bonds and can couple directly with proteins in neurons. A 2024 mouse study reported that 36 THz stimulation suppressed pyramidal-neuron activity by increasing the conductance of voltage-gated potassium channels; another measured roughly 70% of 0.14 THz waves passing through a mouse skull. Human skulls are far thicker and terahertz is strongly absorbed by water, so depth remains the obstacle β but the direction is clear. Interrogation that can calm, suppress or provoke is a different and worse threat than interrogation that can only detect, and the report's "behavioural compliance" prohibition only hints at it.
Every safeguard in the 2026 report assumes a visible device β a headset that is put on, a test that is agreed to or refused. A beamed technology (terahertz or its successors) needs no contact. There is no moment of consent, no moment of refusal, and possibly no awareness it is happening. Consent-based governance has nothing to grip.
The article speaks only of "the brain". The retina and optic nerve are part of the central nervous system and sit behind a transparent window; the spinal cord carries every motor command and sensation. Restrictions written around "brain" devices leave the rest of the CNS β and the peripheral nerves wired into it β open.
Even a flawless detector can't tell whether you know the knife because you used it, or because you saw it on the news, or because an interrogator described it. Better accuracy makes this worse, because the result sounds more authoritative. The article edges toward this β it calls involuntary betrayal of guilty knowledge a "claim" β but never says it outright.
| Issue | 2026 status | 2066 status (technology works) |
|---|---|---|
| Lab vs real-world validity | Strong in the lab (P300 effect size 1.89 in a 2014 meta-analysis), weak in real cases | Resolved β not an objection |
| Countermeasures (suspect deliberately fooling the test) | Older protocols beatable; newer Complex Trial Protocol resists | Resolved for accuracy β but the only defence left is self-manipulation of one's own mind, a mental-integrity issue |
| Altered baselines (drug use, psychiatric conditions) | Known to change P300 amplitude | Resolved for accuracy β but intoxicated or withdrawing people still can't give real consent |
| Recognition β guilt | Central objection | Stronger β accuracy makes wrong answers about guilt more convincing |
| Probe selection and information leakage | Examiner chooses what is tested | Stronger β a perfect detector faithfully confirms whatever police leaked |
| Self-incrimination | US doctrine may not treat a passive brain response as "testimony" | Stronger β a working test is a real alternative to the right to silence |
| NSW loophole | Recognition testing outside the Lie Detectors Act | Live threat, not a hypothetical |
Conclusion: if the technology works, the technical objections fall away and the ethical and legal ones grow. Any argument that depended on failure collapses; any argument grounded in rights holds.
The Neurotransparency Governance Framework (Bill of Cognitive Rights and Neurotechnology Governance) was posted on this forum on 13 June 2026 β four months before the UN Secretary-General's report and five months before the INTERPOL/UNODC report. Thread: https://forum.drugs-and-users.org/index.php?topic=7441.0
Its starting position β the mind as "the final domain of individual sovereignty" β does not depend on the technology failing. That is why it survives the 2066 premise where the article does not.
The framework's central position is that this technology must not proliferate. Its Controlled Access Doctrine sets capability tiers, prohibits public deployment of high-fidelity mental access systems (Tier 4), sets hardware security requirements, and proposes a non-proliferation treaty.
This is the sharpest contrast with the 2026 report. The report regulates use β it accepts the technology will be deployed and tries to ban its coercive applications. The framework prevents availability. Read from 2066, almost every harm in Β§4 and Β§5 β refusal treated as guilt, consumer-device data in court, accidental revelation, neural risk scoring β follows from the technology being widely available. Rules on use can be eroded case by case through judicial discretion, as Β§3.1 shows; a capability that was never distributed cannot be.
Non-proliferation is the only position in either document that addresses the 2066 scenario at its root rather than at its symptoms.
Design points for the treaty:
| Gap in the 2026 article | Framework response |
|---|---|
| Treats deployment as inevitable; regulates use only | Non-proliferation: Tier 4 public deployment prohibited; treaty proposed |
| Ignores technologies that alter neural activity (Β§5.6) | Mental integrity (3) and protection against cognitive manipulation (18) |
| P300 never named | Discusses P300 "brain fingerprinting" directly |
| "Guilty knowledge" left unchallenged | Recognition is not guilt; memories are not recordings |
| Self-incrimination only listed | Neural Fifth Amendment |
| NSW recognition-testing loophole | Mental privacy presumed; burden on whoever seeks access β applies regardless of technique |
| Bans only coercive use | Explicit, granular consent; bundled consent invalid β stricter |
| Consumer-device data route (Β§4.2) | Neural data ownership (6), on-device minimisation (10), limits on government access (16) and commercial exploitation (17) |
| Accidental revelation (Β§5.3) | Inference boundaries (8) β no inferring mental health, criminal propensity etc. without per-category authorisation. In 2066 this becomes the framework's most important principle |
| Commentary not independent of the bodies it reviews | Governance Recursion: "who verifies the verifier" |
Non-proliferation is the framework's first line of defence. The additions below are fallback provisions β they apply wherever containment fails or a capability leaks, so that a breach does not leave people unprotected. Read from 2066, five would close the remaining gaps:
A sixth, structural suggestion: scope the non-proliferation treaty to decoding or modulation capability across the whole CNS, with contactless access in the prohibited tier by default (Β§7.1), and place named techniques in the Controlled Access tiers β P300 recognition testing, BEOS, fMRI (functional magnetic resonance imaging) lie detection, consumer EEG β and terahertz neuromodulation β so no vendor can argue its product belongs in a lower tier.
From 2066, the article reads as an early warning that saw the danger and built its case on the wrong foundation. Its legal analysis β the admissibility discretion and the NSW loophole β was more prescient than its authors likely knew. Its scientific argument was the part with an expiry date.
The greater danger it did not see was never a police officer forcing a headset onto a suspect. It is a world in which the test works, everyone already owns a device that recorded the data, and refusing to be tested is itself taken as evidence of guilt β and, beyond that, a world in which the nervous system can be written to as well as read, at a distance, without contact or awareness. Every safeguard built on consent fails there.
The 2026 report asked how this technology should be used. The more important question is whether it should be available at all. Frameworks grounded in rights rather than in the technology's limits are the ones that last β and a framework that refuses proliferation outright addresses the problem before any courtroom has to.
Prepared with Claude (Anthropic), 18 November 2026.