No, the machine is not yet plucking thoughts out of your skull like grapes.
Good. Breathe.
But do not relax so much that you become livestock.
This is an old story.
Circe did not need to understand every noble sentence in a sailor's mind. She gave him the right cup at the right moment, and down he went into the animal.
The Sirens did not drag the ship by ropes. They sang the song that made men steer themselves onto the rocks.
Māra did not need a sword. He came as fear, desire, doubt, power, glamour, hunger, and the little voice that says, "This is what you really want."
That is the pattern.
So when people say AI can read thoughts, separate the miracle from the theater.
The miracle is real: some people who cannot speak are beginning to speak again through machines that learn the brain's signals for attempted speech. A trapped person gets a bridge back to the world. That is beautiful. Bow to it.
The theater is the headline pretending a machine can hover over any stranger and steal the secret sentence inside his head.
No.
These systems are narrow. Trained. Medical. Often wrong. They hear the door you trained them to hear, not the rooms you keep closed.
But here is where I stop patting your hand.
The real sorcery may not need an implant.
Your phone does not need to read your soul if it can learn your weather.
Tired. Lonely. Angry. Ashamed. Bored. Overloaded. Wanting to disappear.
That is enough.
A system that knows your state can choose the next song.
The next ad. The next outrage. The next comfort. The next little cup from Circe's table.
So the question is not, Can they read my mind?
Too crude.
The question is:
Who is learning the conditions under which I stop being free?
That is where you put your guard.
Not against medicine that gives speech back to the silent.
Against machines that study your weakness and call the leash personalization.
Tie yourself to the mast sometimes.
Use wax sometimes.
Walk without the device sometimes.
Sleep before deciding.
Eat before despairing.
Talk to a living human before believing the private weather in your head.
And when a tool claims to know you, ask one clean question:
Does this return me to myself, or make me easier to steer?
There. That is the line.
Sun is separating three different realities that headlines often collapse into one frightening phrase: medical brain-computer interfaces, experimental language decoding, and commercial state inference.
The first category is real and humane. In trained medical contexts, brain-computer interfaces can record neural activity and use machine-learning systems to translate certain patterns into outputs. As of Stanford's 2023 speech-BCI work, one implanted system decoded attempted speech from a participant with ALS into text at about 62 words per minute after months of training. UCSF's related work helped a paralyzed stroke survivor communicate through a digital avatar. These are not examples of ordinary thought surveillance. They are assistive medical systems designed to restore communication to people whose bodies can no longer express what they intend to say.
The second category is more unsettling but still constrained. Stanford's 2025 inner-speech work showed that imagined speech can produce decodable neural patterns in speech-impaired patients with implanted sensors, but the system was trained, task-specific, and designed to decode only on command. The University of Texas semantic decoder reconstructed the gist of language from fMRI data, but it required cooperation, individual training, and scanner-based measurement; it was not a portable mind-reading device.
So the accurate answer is: no, general-purpose consumer mind reading is not what exists today. What exists is narrower and more conditional: decoding specific neural or physiological signals under specific technical conditions.
The third category may matter most for ordinary life. Companies do not need direct access to private thought if they can infer state from surrounding signals: sleep, movement, heart rate, gaze, scrolling behavior, typing speed, voice, purchasing behavior, location, social context, and timing. Those signals can be used to estimate fatigue, attention, stress, arousal, confusion, or vulnerability. Once a system can estimate state, it can adapt its next prompt, notification, recommendation, ad, or interface.
This is where the institutional boundary begins to appear. The law is learning to name state inference, not just data collection. The FTC has warned that biometric technologies may create privacy, security, deception, bias, discrimination, and unfairness risks. The EU AI Act restricts certain emotion-recognition uses, especially in workplace and educational settings, and prohibits AI practices that manipulate or exploit vulnerability in defined contexts.
But regulation is still behind the commercial primitive. Most legal regimes are better at naming data categories than naming the moment when inferred vulnerability becomes a steering mechanism. A system may not know your private thought and still know enough about your state to time an intervention when resistance is low.
These myths are not magical claims about neuroscience. They are old maps of influence below deliberation - appetite, attention, fear, craving, doubt, and the pull of false necessity.
The ethical boundary is therefore not brain technology good or brain technology bad. The boundary is whether the system restores agency or extracts agency under conditions of vulnerability.
A speech neuroprosthesis that helps a paralyzed person communicate restores agency. A platform that detects loneliness or fatigue and uses that state to intensify compulsion extracts agency.
The practical response is not panic. It is discernment, law, design, and personal practice. At the individual level, protect moments of choice when your state is compromised. At the product level, refuse designs that use inferred vulnerability as a conversion opportunity. At the social level, treat neural, biometric, and affective data as unusually sensitive because they reveal not only what a person does, but when a person is easiest to influence.
In institutional language, Sun's test becomes: agency restored, or agency extracted.