BCI Security Plugin
Open-source security toolkit for AI coding platforms. Scans BCI code, detects PII in neural data, generates compliance reports, and enforces neuroethics guardrails. Works with any AI agent that can read files and follow structured prompts.
Test It Today
No BCI device needed. The plugin ships with pre-tagged sample data from open research.
Install
git clone https://github.com/qinnovates/bci-security.gitClaude Code:
claude plugins install --scope user ./bci-securityOther AI platforms: The plugin is pure markdown and JSON. No compiled code, no dependencies. Copy the skills/ directory contents into your platform's skill format, or paste SKILL.md instructions directly into your AI conversation with the relevant data JSON.
/bci-scan --demo Scan sample ADHD research data. See all 7 detection rules catch transport, PII, credential, ML, and stimulation issues.
/bci anonymize --demo Scan neural data files for PII in headers, filenames, and metadata. Get anonymization scripts.
/bci compliance --demo Map findings to GDPR, CCPA, Chile Neurorights, UNESCO, Mind Act, and HIPAA.
/bci explain QIF-T0001 Look up any of 161 TARA threat techniques with three layers of detail.
Capabilities
Transport encryption, data storage PII, API credentials, regulatory PII, ML model security, stimulation safety bounds, MNE/NWB pipeline security
Scans EDF, BDF, XDF, FIF, NWB, GDF, CSV, MAT files for PII. Checks filenames, headers, subject metadata. Generates pyedflib, MNE, and pynwb remediation scripts.
9 compliance domains mapped to GDPR, CCPA, Chile Neurorights, UNESCO Recommendation, Mind Act, and HIPAA. 18 PII detection patterns with remediation roadmaps.
Full threat catalog with NISS severity scoring (6 neural-specific dimensions), evidence tiers, therapeutic analogs, and defensive controls per QIF hourglass band.
Guardrails (8 neuroethics constraints, regulatory mappings, status qualifiers) + Hardening (10 credential regex patterns, prompt injection defense, report sanitization, consent gate, pre-commit hook).
ADHD vs ASD: EEG Differences for Threat Modeling
Different neurodevelopmental conditions produce different signal environments. TARA techniques affect ADHD and ASD users differently. The sample data includes both conditions so threat models account for population-specific risk profiles. These are EEG pattern references for threat modeling purposes, not diagnostic markers.
| EEG Marker | ADHD | ASD |
|---|---|---|
| Gamma at rest | Nothing notable | Elevated (g=0.37) — strongest ASD-specific power marker |
| Theta at Cz | Elevated in subgroup | No deviation |
| Alpha | Coherence disrupted | Power reduced globally |
| P300 latency | Prolonged | Normal (amplitude reduced only) |
| N170 face ERP | Normal (gaze-specific only) | Broadly delayed — most replicated ASD ERP finding |
| Connectivity | Alpha-band frontal disruption | Long-range underconnectivity across multiple bands |
A TARA technique operating in gamma at rest hits an elevated baseline in ASD but not ADHD. A technique timed to P300 windows hits delayed responses in ADHD but normal timing in ASD. Coherence patterns are structurally different: ADHD disrupts alpha frontal networks, ASD disrupts broader temporal-frontal pathways.
Sample EEG Data
Qinnovate compiled and pre-tagged these samples from published open-access research. All datasets are anonymized metadata and configuration only. No raw EEG signal data is distributed through the plugin.
| Dataset | Subjects | Channels | DSM-5 | License |
|---|---|---|---|---|
| ADHD Adult Resting State | 79 (37 ADHD + 42 controls) | 5 | F90.0 | CC BY 4.0 |
| ADHD Children | 121 (61 ADHD + 60 controls) | 19 | F90.0 | Academic |
| ADHD Focus + Gameplay | ADHD during tasks | 14 | F90.0 | Academic |
Explore all 16 datasets in the EEG Data Studio. Filter by condition (ADHD, epilepsy, motor imagery, emotion), type (real, synthetic, simulated attack), or search by DSM-5 code.
Device Integration
The plugin scans code that connects to devices. It does not connect to devices directly. Community-supported integrations based on open-source SDKs. QInnovate is not affiliated with any device manufacturer.
Cyton, Ganglion, Daisy
20+ board support
Research analysis
DANDI Archive
CortexPy API
Lab Streaming Layer
Security notice: Before connecting any BCI device to an AI-assisted analysis pipeline, ensure your network is secure and segmented. Have security experts review your integration and compliance experts review regulatory requirements. If your institution is using an AI coding platform for BCI research, they have likely already performed the necessary security reviews and compliance checks for that platform's data processing agreement. See the full Integration Guide for details.
Security Hardrails
Hardrails = guardrails (ethical constraints) + hardening (technical enforcement). Both layers apply to every scan, report, and assessment.
- 8 neuroethics constraints from published literature
- 6 regulatory frameworks (GDPR, CCPA, Chile, UNESCO, Mind Act, HIPAA)
- Status qualifiers on all QIF references
- Dual-use framing: every threat paired with defense
- 10 credential regex patterns (zero-tolerance redaction)
- 17-keyword prompt injection defense
- 7-rule report sanitization with self-verification
- Neural data consent gate
- Pre-commit hook blocking secrets and session artifacts
What This Is and What It Isn't
This plugin is designed for reviewing anonymized BCI data with AI coding agents. It performs clinical and threat mapping based on patterns that research labs derive from their own data. Qinnovate provides pre-tagged samples from open research so you can test. The plugin is modular: use one skill or all eight.
- Research tool, not a medical device
- Not legal advice — compliance mappings are simplified for threat modeling
- "No issues detected" does not mean "compliant"
- All findings require independent verification by qualified professionals
- TARA and NISS are proposed frameworks, not adopted standards
Documentation
Installation, commands, quick start
Security specification, trust model, credential patterns
Component map, data flow, hardrails model
OpenBCI, BrainFlow, MNE, NWB, Emotiv, LSL
Legal notices, privacy, liability limitations
Browse 16 EEG datasets, filter by condition