The fundamental commitment of the Brainstorm Reactor is isolation. Your strategic data is your own.
As the operational impact of AI increases, regulatory frameworks like Regulation (EU) 2024/1689 (European AI Act) demand transparency, consent, and strict definitions of data handling. The Brainstorm Reactor is designed to natively comply with these requirements through structural data abstinence and automated provenance injection.
Intellectual Property Protection
The engine does not use user inputs, session data, or generated outputs to train, fine-tune, or calibrate any underlying models.
Interactions with external providers operate exclusively through Zero-Data-Retention (ZDR) endpoints. This guarantees that once a request finishes executing and returns to the Next.js edge, the payload is immediately dropped by the provider system. It is never stored in external memory banks or utilized for future model generations.
Machine-Readable Provenance & Article 50 Compliance
In full accordance with Article 50(2) of the EU AI Act (enforced August 2, 2026), the Brainstorm Reactor embeds machine-readable provenance metadata into synthetic text, media, and export files:
- Web DOM Metadata: The application dynamically injects Schema.org
AIAnnotationsJSON-LD structured data into the document<head>and tags message containers with global WHATWGai-disclosure="ai-generated"attributes aligned with the IPTCDigitalSourceTypetaxonomy (http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia). - PDF & Document Exports: Exported PDFs (e.g., LinkedIn Carousels) include embedded C2PA v2.4 (ISO 22144) XMP metadata assertions (
c2pa.actions: c2pa.created), compiled via server-side SHA-256 byte-range exclusion hashing without triggering rendering overhead. - Media Pass-Through: Generated image and audio assets maintain native EXIF, IPTC, and SynthID/C2PA header chunks across all proxy routes.
Transparent Usage & Human-in-the-Loop (Art. 14 & 26)
Under Articles 14 and 26 of the AI Act, human oversight and operational transparency are mandatory:
- Human-in-the-Loop Governance: Every workflow recipe requires explicit user steering at each step. The system executes no autonomous actions without user consent.
- Transparent Billing: Costs are handled transparently via a credit-based formula applied strictly per execution run, completely decoupling operational costs from obscure "token counting" dynamics.
- Ephemeral State: Users retain absolute agency over saved session data (The Pool) versus data discarded upon session completion.
Audio Intelligence and Biometric Exclusion
The live transcription engine processes speech-to-text exclusively for content extraction. The system explicitly excludes biometric identification, emotional state inference, and voice profiling — prohibited categories under Art. 5 of Regulation (EU) 2024/1689. Speaker diarization operates purely on acoustic signal separation and does not attempt identity resolution.
Raw audio data is processed as a transient stream and discarded immediately after transcription. No audio recordings are stored on the platform's servers.