1. Overview

Brainpercent is an AI-powered content generation tool. This page explains how the AI works, what it can and cannot reliably do, and your responsibilities when using AI-generated outputs. We publish this page in fulfillment of Article 4 (AI literacy) of the EU AI Act (Regulation 2024/1689) and to document, honestly, where we stand today on the Article 50 (transparency) obligations, including the work still in progress.

2. How It Works

When you submit a URL, prompt, or brand asset, Brainpercent orchestrates a series of AI model calls to produce content. The flow typically involves:

Scraping the source URL or analyzing the uploaded asset (no AI model — deterministic extraction)
Planning what to generate (large language model selects format, tone, structure)
Generating text, images, video frames, or audio (specialized models per modality)
Quality checks a second AI pass to detect fabricated facts in source-grounded content
Assembly into the final output you see (deterministic post-processing)

Your prompts are sent to third-party AI providers. We never use your prompts or outputs to train any model of ours, and our agreements with those providers prohibit training on customer data.

3. AI Models We Use

We use a mix of frontier AI models from leading providers. The specific model selected depends on the task, content type, and language. Current providers include:

Anthropic Claude — text generation, fact-checking, classification
OpenAI (indirect only) — reached indirectly only: Sora video via the GoAPI gateway and the voice assistant language model via Vapi. No direct OpenAI API integration since June 2026
Google VEO, OpenAI Sora, Kling AI, Seedance — video generation
Flux, Gemini, nano-banana — image generation
GoAPI, Cloudinary — voice and audio

We may change models or providers without notice. See /legal/subprocessors for the current list.

4. Known Limitations

AI models — including the most capable frontier models — have well-documented limitations:

Hallucinations: models can produce plausible-sounding but factually incorrect statements, fabricated quotes, made-up statistics, and invented citations.
Training cutoff: models have a knowledge cutoff date and do not know about events after that date.
Reasoning errors: mathematical calculations, multi-step logic, and chronological reasoning can be unreliable.
Bias: models may reflect biases present in training data, including gender, race, cultural, and political biases.
Inconsistency: the same prompt can produce different outputs on different runs.
Similar outputs: different users with similar prompts can receive similar or identical outputs.
Style drift: over long outputs, tone and voice may shift mid-piece.

5. How to Verify Outputs

Treat AI outputs as a first draft, not a finished product. Before publishing or relying on any AI output:

Verify every fact — names, dates, statistics, citations, quotes. Search for the original source.
Check claims against authoritative sources (official websites, original reporting, government data).
Review for bias and tone — does it represent the subject fairly?
Confirm brand alignment — does it match your voice and values?
For images/video of people — confirm consent has been obtained for any real person depicted.
For regulated industries (legal, medical, financial, safety) — have a qualified professional review before publication.

6. Labeling AI Content

EU AI Act Article 50 (obligations enforceable from 2 August 2026) and many national regulations require that AI-generated content be identifiable as such. Here is where Brainpercent stands, stated plainly:

What we do today: you are informed that you are interacting with an AI assistant, all content in your workspace is AI-generated by design and presented as drafts for you to review, and the platform reminds you to review content before publishing. This transparency page documents how our AI works, and our AI assistants disclose that they are AI and will truthfully say so if you ask.
In progress (target: 2 December 2026): embedding machine-readable provenance (C2PA Content Credentials) into generated images and videos, plus a free verification route where anyone can check whether a file was generated by Brainpercent. Until this ships, our outputs carry no machine-readable marking.
Your duties as a deployer (Article 50(4)): when you publish AI-generated content, you must label deepfakes (synthetic content depicting real persons, objects, places, or events that appears authentic), and you must disclose AI-generated text published to inform the public on matters of public interest, unless that text underwent human editorial review and a person holds editorial responsibility for it.
Other jurisdictions: the US FTC (16 C.F.R. Part 255) requires disclosure of AI-generated endorsements and testimonials in commercial contexts; China, South Korea, and Singapore have AI labeling requirements; check your local law and each platform's AI-labeling rules.

To be explicit: we do not claim full Article 50 compliance today. Human-readable labeling exists now; machine-readable marking is scheduled before the 2 December 2026 grace deadline for that obligation. Until then, and after, you remain responsible for the deployer disclosures above whenever you publish AI-generated content externally.

7. High-Risk Uses (Prohibited)

Brainpercent is not classified as a high-risk AI system under Annex III of the EU AI Act. You may NOT use the Service for:

Decisions affecting fundamental rights, employment, education access, or social benefits
Biometric identification or categorization of natural persons
Critical infrastructure management
Law enforcement or judicial decision-making
Migration, asylum, or border control assessments
Medical diagnosis, treatment planning, or safety-critical decisions

Such uses fall under the EU AI Act high-risk system regime, which imposes substantial compliance obligations (risk management system, data governance, transparency, human oversight, accuracy + robustness). Brainpercent is not designed or certified for these uses.

8. Contact

Questions about AI transparency or model behavior: edward@brainpercent.com with subject "AI Transparency".