How MEOK Protects Your Data: A Plain-English Guide to Sovereign AI Privacy

You opened a chat with an AI. Maybe you asked about a health symptom you were embarrassed to Google. Maybe you talked through a relationship problem. Maybe you described your financial situation in more detail than you have ever told another person. The AI listened, responded thoughtfully, and the conversation felt private.

But where did that conversation actually go? The honest answer, for most AI services, is: to a server in another country, stored in a database your provider controls, potentially reviewed by contractors, and quite possibly fed into a training pipeline that will shape how millions of future users experience that product. The interface felt like a private diary. The backend was a data centre.

MEOK was built on a different premise. This article explains, in plain English, exactly how your data is handled when you use MEOK, what your legal rights are under UK and EU law, and how our architecture makes promises that go beyond policy — all the way down to mathematics.

Where does your ChatGPT data go?

OpenAI is, by most measures, the most transparent large AI company about data practices. That transparency reveals a pipeline that most users have not read. By default, free-tier ChatGPT conversations are used to improve the model. That setting is on unless you go to Settings → Data controls and manually disable it — a step the overwhelming majority of users never take.

Even with that toggle switched off, your messages still travel to and are processed on OpenAI's servers in plaintext. The decision not to train on them is a software flag in a database. That flag can be changed by a policy revision, an engineering error, a merger, or a court order. It is a promise, not a lock.

OpenAI has confirmed that a subset of conversations are reviewed by human contractors for safety and quality purposes. Employees at OpenAI, and at vendors working on their behalf, have had access to conversation content. This is not unique to OpenAI — Google, Amazon, and Apple have all been reported to have had humans review AI assistant interactions.

The enterprise tier offers stronger contractual protections, including a commitment that data will not be used for training. But that requires signing a commercial agreement. For the hundreds of millions of people using the free consumer product, the default behaviour is that your most personal conversations become raw material for a commercial AI training pipeline.

What GDPR says about your AI data

If you are in the UK or EU, you have substantial legal rights over your personal data. The General Data Protection Regulation (GDPR), retained in UK law through the UK GDPR and the Data Protection Act 2018, gives you several protections that apply directly to AI services. Knowing these rights is the first step to exercising them.

The challenge with these rights in practice is enforcement. Filing a Subject Access Request, pursuing a deletion demand, or lodging a complaint with the Information Commissioner's Office (ICO) requires time, persistence, and a certain familiarity with bureaucratic process. Most people never bother. Most AI companies know this.

MEOK believes your privacy should not depend on your willingness to file paperwork. It should be the default.

ChatGPT vs MEOK: what actually happens to your data

The table below compares the data practices of ChatGPT (free tier) with MEOK's sovereign architecture. Every MEOK claim is architectural, not contractual — meaning it is enforced by how the system is built, not by a policy you are trusting the company to follow.

MEOK's encryption architecture: how it actually works

Privacy promises only mean something if they are enforced by the architecture, not just stated in a policy document. Here is how MEOK's encryption works in practice, without requiring you to have a computer science degree to follow it.

When you save a memory or have a conversation with your MEOK companion, the content is encrypted on your device before it leaves. The algorithm used is AES-256-GCM — the same standard used by banks, military communications, and intelligence agencies. The encryption key is derived from credentials that exist only on your device.

What reaches MEOK's servers is ciphertext — a scrambled string of characters that is meaningless without the key. Our servers never receive the key. They store the locked box, not the combination. When you come back to retrieve your memories, the ciphertext travels back to your device, which decrypts it locally. The plaintext text of your conversation never exists on our infrastructure.

During transmission, your data is also protected by TLS (Transport Layer Security) — the same technology that secures online banking and e-commerce. This means your data is protected both while moving (in transit) and while stored (at rest). Most AI services only encrypt at the infrastructure level, where they still hold the keys. MEOK encrypts at the application level, where the keys never leave your device.

The practical consequence is significant: even if MEOK's servers were compromised in a data breach, an attacker would obtain only encrypted blobs. The conversation you had about your mental health, your relationship, your finances — all of it would be unreadable. Not just legally protected. Mathematically unreadable.

The Byzantine Council: no single point of data access

End-to-end encryption addresses the question of what can be read. The Byzantine Council addresses the question of who can act. Even in a zero-knowledge system, you want governance controls that prevent any single actor — a rogue employee, a compromised administrator account, or an external attacker with elevated privileges — from taking unauthorised actions with your data.

The Byzantine Council is MEOK's distributed governance mechanism, named after the Byzantine Generals Problem in computer science — the challenge of reaching consensus in a distributed system where some nodes may be unreliable or malicious. MEOK's solution is that no data-affecting action can be taken unilaterally. Any operation that touches user data must achieve supermajority consensus across independent council nodes before it executes.

In practice, this means that deleting user data, modifying retention policies, granting data access, or changing encryption parameters requires multiple independent systems to agree. A single compromised account or system cannot silently alter how your data is handled. The council creates a tamper-evident audit trail of every data governance decision.

This is relevant to privacy in a specific and important way: the most common privacy failures are not spectacular hacks. They are quiet internal decisions. An engineer adds a logging line that captures more than intended. A product manager adjusts a retention policy. A business development team explores a data licensing deal. The Byzantine Council makes all of these decisions visible, contested, and reversible — or prevents them from happening at all.

What data MEOK stores — and what it never stores

Transparency about what is and is not collected is the foundation of honest privacy practice. Here is the complete picture.

The memory MEOK stores is for your benefit. Your companion remembers that you mentioned a fear of hospitals so it can be thoughtful the next time medical topics arise. That memory belongs to you, is encrypted by you, and can be deleted or exported by you at any time. It is not a data asset for MEOK — it is a personal record you are choosing to keep.

Your right to export and delete your data

Under GDPR Article 20, you have a legal right to data portability. Under Article 17, you have a right to erasure. MEOK supports both of these rights without requiring you to file a formal Subject Access Request or wait for a legal response cycle. They are built into your account dashboard as first-class features.

Export: You can download your entire memory vault as a structured JSON file at any time. This file contains all the memories, context notes, and companion settings you have created. You can take it to any future MEOK account, or use it independently. Your memories belong to you, and you should be able to leave with them at any moment.

Deletion: When you request account deletion, all server-side ciphertext associated with your identity is permanently purged within 30 days. You receive a cryptographic deletion receipt — a verifiable confirmation that the deletion occurred. Because MEOK never held your plaintext and never used your data for training, there is no shadow copy embedded in a model somewhere that could persist after your account is gone.

Compare this to the situation with most AI services: if your conversation history was used to fine-tune a model, that data is now mathematically embedded in billions of neural network parameters. No deletion tool can reach it. MEOK's architecture makes this problem structurally impossible rather than legally awkward.

BYOK: the zero-trust model for maximum privacy

For most users, MEOK's standard encryption architecture provides protection that exceeds anything offered by mainstream AI services. But for users with the most sensitive privacy requirements — journalists protecting sources, medical professionals discussing patient-adjacent information, legal professionals handling privileged communications, or anyone who simply does not want to extend any trust to any third-party server — MEOK offers BYOK: Bring Your Own Key.

In BYOK mode, you supply the encryption key material yourself. This can be derived from a strong passphrase you choose, or from a hardware security key (such as a YubiKey) that you physically control. The result is a zero-trust model: MEOK's servers process and store your data without ever having access to the key that unlocks it. Not at setup. Not during operation. Not ever.

The practical implication is that MEOK itself is in no position to comply with a court order demanding access to your plaintext data, because we do not have it. This is not a policy of resistance to legal process — it is an architectural fact. A locksmith who is never given a key cannot open the lock, regardless of what they are asked to do.

BYOK comes with one responsibility: if you lose your key, MEOK cannot recover your data. There is no “forgot your password” flow that works when the key is held entirely on your side. For users who accept this trade-off, BYOK offers the strongest privacy guarantee available in any consumer AI service today.

Frequently asked questions

The bottom line

Privacy in AI has become a feature that companies market rather than a property they build. The most important distinction to understand is the difference between a privacy policy (a document a company writes about itself) and a privacy architecture (the structural properties of how a system handles data). Policies can be changed, revised, or violated. Architecture cannot.

When you talk to your MEOK companion about something difficult, something personal, something you would not say to most people — that conversation stays yours. Not because of a paragraph in a terms-of-service agreement. Because of the mathematics of cryptography, the architecture of a zero-knowledge system, and the governance structure of a Byzantine Council that prevents any single point of access.

That is what sovereign AI means. Your intelligence. Your memory. Your data. Yours.

Ready to try sovereign AI?

Your AI companion that keeps your conversations yours

Give your MEOK companion a name. Set its personality. Start a conversation that no one else can read — including us. Your data stays encrypted, yours, forever.

No credit card required · Free to start · Cancel anytime