We didn't hack the password.
Multi-modal behavioral obfuscation engine that protects your cognitive fingerprint from surveillance systems.
Scroll to explore
An active defense system that weaponizes gradient starvation to establish a technically enforceable Right to Be Inscrutable.
Modern surveillance infrastructures now target the high-fidelity behavioral telemetry underlying human-computer interaction: mouse movements, keystroke dynamics, gaze patterns, and increasingly, direct neural signals from brain-computer interfaces.
Cognitive Canary exploits a structural vulnerability in machine learning — gradient starvation — to establish not a plea for better classifier behavior, but a technical enforcement of privacy, akin to encryption layers provided by Tor or Signal.
| Category | Metrics | Mathematical Basis | Inference Potential |
|---|---|---|---|
| Kinematic | Velocity, Acceleration, Jerk | Time derivatives of displacement | Motor skill, fatigue, aging |
| Geometric | Curvature, Path Efficiency | Path-to-Euclidean ratio | Intention, confusion, spatial ability |
| Spectral | PSD, Dominant Frequency | Fourier Transform | Essential tremor, anxiety, stress |
| Entropic | Spectral Entropy, Shannon | Spectrum irregularity | Human vs. Bot, cognitive load |
Brain-computer interfaces record high-bandwidth neural data encoding far beyond motor intent: attention states, emotional valence, elements of intended speech, and decision-making processes.
1,024 channel intracortical • PRIME Study • R1 robot insertion
Apple BCI HID integration → iPhone/iPad/Vision Pro • COMMAND study
Minimally invasive • Speech & motor recovery • High resolution
Cognitive Canary leverages gradient starvation (Pezeshki et al., 2021) — a fundamental learning bias where over-parameterized neural networks capture only the "cheapest" predictive features while ignoring others.
Like Clever Hans, the horse that appeared to do arithmetic but responded to subtle cues, neural networks rely on shortcuts rather than robust features.
Canary intentionally induces gradient starvation by flooding the feature space with synthetic but biologically plausible noise. By making "cheap" spectral features statistically meaningless, Canary forces classifiers into a regime where they fail to develop representations for sensitive neuromotor signatures.
Injects small-amplitude Lissajous curves into cursor data using superposition of perpendicular simple harmonic motions.
Adds low-variance Gaussian jitter with 1/f spectral structure — a hallmark of biological feedforward control.
Adjusts injection amplitude based on task sensitivity. High injection during passive browsing, moderate during text editing.
Zero-knowledge proofs ensure the defense mechanism doesn't become a surveillance sink.
| Task Type | Injection | Rationale |
|---|---|---|
| Passive Browsing | High | High surveillance stakes; low precision stakes |
| Text Editing | Moderate | Balance obfuscation and character selection |
| Precision Design | Low / Disabled | Fine control critical; usability priority |
| Gaming / AR | Adaptive | Minimized latency; rapid response preservation |
3 humans implanted • 1,024 ch intracortical
Apple BCI HID integration • COMMAND study
62 WPM speech restoration • FDA Breakthrough
EEG signals can re-identify individuals even from consumer wearables. P300, ERN, N200 latency profiles are permanent, uncancellable fingerprints.
Synchron's BCI HID profile on iOS marks the formal canonization of neural input. With thought-control now real, behavioral telemetry pipelines gain a direct cortical feed.
Current connectome models miss astrocyte regulation entirely. Neurofeedback products target alertness and mood via EEG proxies with no regulatory framework.
Core obfuscation engine with 2D Lissajous cursor rewriting.
EEG Shield (3-layer differential privacy), Neuro Audit, 7-engine suite complete.
GAN-based behavioral synthesis, OS-level driver integration, WebExtension.
Full neuro-sovereignty stack: cross-device identity unlinking, regulatory API bridge.