General anesthesia: how consciousness goes out, and how it comes back

Sep 17, 2026 · #neuroscience #consciousness #anesthesia #brain

Anesthesia is the only routine intervention that removes consciousness on schedule and reliably gives it back, which makes it the best experimental handle the science of consciousness has. It is also a standing demonstration of how far control can run ahead of understanding.

An article card showing a brain rhythm that is fast and irregular while awake, slow and regular under general anesthesia, and fast again during emergence

An anesthetist pushes a syringe of white emulsion into a vein. Within about thirty seconds the person on the table stops answering, their eyelids stop responding to touch, and for the next several hours a surgeon can open their abdomen without them minding. Then the infusion stops, and a few minutes later the person comes back. Not a copy of them. Them, with their memories, their irritations, their sense of being someone in particular. This happens hundreds of millions of times a year, it is routine enough to be boring, and nobody can tell you precisely what is being switched off.

That is not a rhetorical flourish. The molecular end of the story is in decent shape. Propofol, the white emulsion, works mainly on the GABA-A receptor: a channel in the neuron’s membrane that opens when the brain’s main inhibitory messenger lands on it, lets chloride ions in, and makes the cell harder to fire. Propofol holds that channel open longer, so every inhibitory signal lands harder. Inhaled agents such as sevoflurane hit overlapping targets. A recent review of induction mechanisms frames the whole drug class as substituting for the brain’s own messengers and hijacking the circuitry that normally runs the switch between sleep and waking.

All of which explains very little about consciousness. Knowing that a drug potentiates inhibition across the cortex tells you where the molecule sits. It does not tell you which property of a brain has to fail before there stops being anyone home. The distance between those two statements is the whole subject.

Unresponsive is not the same as unconscious

Start with the measurement, because the measurement is where the trouble is.

In an operating theatre, “asleep” is inferred from behaviour: the patient does not move, does not answer, and afterwards does not remember. Each of those is a proxy, and the standard surgical recipe deliberately breaks the first one. Neuromuscular block, the paralytic given so the surgical field stays still, disconnects muscle from nerve. A paralysed patient cannot report anything, whatever is going on inside.

The field has one clean trick for this, and it is beautifully low-tech. Before the paralytic goes in, a blood pressure cuff on one upper arm is inflated above systolic pressure. The drug never reaches that forearm, which stays capable of movement while the rest of the body is still. Then, during surgery, somebody asks the patient to squeeze their hand. This is the isolated forearm technique, and it remains the closest thing anyone has to asking the question directly.

The answers are uncomfortable. A meta-analysis pooling twenty two studies and 1131 patients found that 34.8 per cent of patients responded to command at some point during induction or maintenance, with 31.2 per cent responding during maintenance, meaning during the surgery itself, at doses considered adequate. Most of them have no memory of it afterwards.

Set that beside the figure for awareness as patients report it. NAP5, a UK and Ireland audit that canvassed every public hospital, put spontaneously reported accidental awareness at roughly one case in 19,600 anesthetics. Split by technique the number moves hard: about one in 8,200 when a paralytic was used, about one in 136,000 when it was not. Two thirds of the reports came from the dynamic phases, induction and emergence, rather than from the long stable middle.

Put the two numbers together and the shape of the problem appears. Awareness that a patient remembers and reports is genuinely rare. Responsiveness to a question, under anesthesia deep enough for surgery, is not rare at all. They measure different things, and neither of them measures experience.

The working vocabulary reflects that split. Connected consciousness means experience plus contact with the outside world. Disconnected consciousness means experience with the outside world sealed off, which is what an ordinary dream is. Unconsciousness proper means neither. A patient can be in any of the three while looking identical from the far side of the drapes.

What the brain looks like on the way out

If behaviour is a poor readout, the alternative is to watch the brain directly. The oldest signature is also the most robust: as propofol takes hold, the EEG, the summed electrical rhythm of the cortex picked up by scalp electrodes, develops a large slow oscillation and a strong rhythm near ten cycles per second that migrates toward the front of the head. Anesthetists call it anteriorisation, and it is reliable enough to steer by.

What generates it is a loop. The thalamus, a pair of egg-shaped structures in the middle of the brain, is the switchboard through which nearly all sensory traffic and a great deal of cortex-to-cortex traffic passes. Its neurons come in two broad flavours: core cells that project to one specific cortical address, and matrix cells that spray diffusely across the outer layer of large stretches of cortex. A 2024 human imaging study found that deep propofol sedation selectively degrades that core and matrix architecture, flattening the functional geometry that normally separates sensory regions from the association areas sitting on top of them. A 2025 human recording study went further, reporting that high-order thalamic nuclei gate whether a stimulus is consciously perceived at all, through a loop running out to frontal cortex and back. A 2026 widefield imaging study in mice found that different anesthetics leave distinctly different whole-cortex signatures rather than one common shutdown.

The convergent claim is not that the cortex switches off. Under propofol the cortex is metabolically busy and auditory areas still respond to sound. The claim is that cortex and thalamus stop exchanging signals in a way that carries information forward, and that this, rather than silence, is what unconsciousness looks like from the outside.

The complexity test, and the drug that breaks the pattern

The most direct probe available skips passive observation entirely. Fire a single magnetic pulse through the skull with transcranial magnetic stimulation, which makes a patch of cortex discharge, then record on EEG what the rest of the brain does with the disturbance. Awake, the response spreads, reverberates through distant regions and takes a different shape each time. The perturbational complexity index compresses that reply into one number, on the reasoning that a conscious system should both integrate the disturbance across regions and respond to it in a differentiated way.

The experiment that matters compared three anesthetics at doses where every participant was unresponsive. Under propofol the pulse produced a local response that died roughly where it started. Under xenon it produced a global response that was stereotyped and simple. Under ketamine it produced a complex, spreading pattern, and those participants woke up reporting vivid dreams. The propofol and xenon groups reported nothing at all.

That is the cleanest dissociation in the literature. Three drugs, one behavioural state, and a physiological measure that sorted them by whether there was any experience rather than by whether there was any movement. It also means unconsciousness is not one thing that anesthetics do: ketamine takes a different route and arrives somewhere else.

Perturbational complexity is a research instrument, not a monitor. Nobody is wheeling a magnetic stimulator into every theatre. A 2024 study found that signatures of criticality in ordinary resting EEG, meaning a regime poised between order and randomness in which a disturbance neither dies immediately nor runs away, predicted both anesthetic loss of consciousness and the complexity index itself. If that holds up, the expensive measurement may turn out to have a cheap proxy.

Coming back is not the drug leaving

The textbook account of waking up is subtraction: the drug washes out, the brain resumes. That account is wrong, or at least badly incomplete.

The dose at which a subject goes under is not the dose at which they come back. The system shows hysteresis, a resistance to changing state in either direction, which the field calls neural inertia. Emergence behaves like an active reassembly with circuitry of its own, not a passive drain.

The sharpest recent evidence comes from the rarest kind of recording. In 2026 a group reported single neuron activity in humans tracked continuously through emergence, in neurosurgical patients who already had depth electrodes implanted for clinical reasons. Firing rates rose as emergence proceeded. Activity changed in some regions before any response to command, earliest in the hippocampus, the parahippocampal gyrus and the amygdala, then cingulate cortex, then insula. The population as a whole drifted toward criticality as the person came back, having sat in more rigid, attractor-like states under the drug.

The caveats are real and worth stating plainly. It is a preprint at the time of writing. The sample is small, and a brain with electrodes in it is there for a reason, which is not a neutral starting point. And activity that precedes a response is not thereby the cause of it: the medial temporal lobe is a memory structure, so early reactivation there may be the brain reloading its context rather than relighting awareness.

What still breaks

Four things, and none of them are minor.

The monitors are weak. The bispectral index and its competitors reduce the EEG to a single number between 0 and 100 through proprietary algorithms. Patients who respond on the isolated forearm frequently sit inside the range labelled adequate. And the clinical promise that would have justified the hardware, fewer cases of postoperative delirium, has not held up. ENGAGES-Canada randomised 1140 cardiac surgery patients aged 60 and over, cut time in EEG suppression by 66 per cent in the guided arm, and found delirium in 18.15 per cent of that arm against 18.10 per cent of usual care.

The validation is circular. Every depth-of-anesthesia monitor is trained and validated against behaviour and later recall, and behaviour and recall are precisely the things that come apart from experience. There is no ground truth to calibrate against, which is not an engineering gap but a conceptual one.

Most of the circuit-level mechanism is rodent work. Cortical dynamics, brainstem pathways and the arousal nuclei implicated in emergence are mapped in mice; the human evidence is largely correlational imaging plus a handful of rare electrode studies.

And the drugs disagree with each other. If propofol, xenon and ketamine reach clinical unresponsiveness by different dynamics, there may be no single switch to find, only several different ways of interrupting whatever an awake brain is doing.

The uncomfortable summary

Anesthesia is the only routine intervention that removes consciousness on schedule and gives it back. That makes it the best experimental handle the science of consciousness has, better than sleep, which cannot be dialled, and better than brain injury, which cannot be reversed. It is also a permanent reminder of how far control can run ahead of understanding. An engineer looking at the situation would say we have an excellent controller and no model.

What the last decade has actually established is narrower than the headlines and more interesting than them. Being unresponsive and being absent are different states. They can be told apart physiologically. And the measure that tells them apart is not about how much activity a brain has, but about how richly it answers back when you poke it. Whether that property is the thing itself or merely its most reliable shadow is still open. So is the question underneath it: whether what stops under propofol is a computation that could in principle run on other hardware, or a state that a particular kind of wet, noisy, self-correcting tissue has to be in.