The server cannot read your files. Even under court order.

What "zero-knowledge" really means

The term is overloaded in marketing. In SkyeConnex's specific implementation, zero-knowledge is enforced at three layers simultaneously:

1. Cryptographic. Every file has a unique data-encryption key (DEK), wrapped by a User Master Key (UMK) that the server never holds in unwrapped form. The UMK is derived from your password or recovery key — both of which only you know.
2. Topological. Files are Reed-Solomon split into seven shards across multiple providers. Even if the SkyeConnex server were fully compromised, no shard on its own contains plaintext, and no five shards land on a single provider.
3. Jurisdictional. Shards span the countries you allow. Compelling decryption requires coordinated legal action across multiple jurisdictions and a working break against AES-256-GCM + ML-KEM-1024.

Why hyperscaler "customer-managed keys" are not zero-knowledge

When AWS, Azure, or GCP describe customer-managed keys (CMK), the key is held in the provider's KMS — meaning the provider's plane has operational access to it whenever the customer's workload requests a read. The provider can be served a court order that requires using those keys, and the unwrap happens within the provider's environment.

SkyeConnex is structurally different. The UMK is derived client-side and never transmitted in unwrapped form. The server simply does not have the capability to decrypt — there is no API call that returns plaintext.

What about password loss?

SkyeConnex generates an out-of-band 24-character recovery key during account setup. That key independently wraps the UMK — so you can lose your password without losing your files, and zero-knowledge is preserved throughout because the recovery key never leaves your custody.

For organisations that need higher assurance, Split-Authority Decryption uses Shamir Secret Sharing to require an M-of-N custodian quorum to unlock the most sensitive scopes — eliminating the single-point-of-recovery risk entirely.

Duress mode

Under coercion, a user can authenticate with a secondary "duress password" that unwraps a limited shadow scope — preserving operational secrecy without revealing that duress mode is in effect.

How does this affect performance?

Client-side encryption adds a fixed CPU cost on the device but eliminates many round-trips, because authorisation happens at upload — not on every read. Frame-level AES-GCM is hardware-accelerated on every platform we ship: web (WebCrypto), mobile (iOS/Android crypto APIs), Windows virtual drive (CNG). Memory cost is one frame (~5 MB) regardless of file size — even multi-gigabyte downloads stream.

Tier-change neutrality

Subscription tier changes — upgrade, downgrade, or downgrade-and-re-upgrade — never affect recoverability. An organisation that leaves the Sovereign tier retains access to every file it encrypted while on it. By construction, not by support ticket.

What zero-knowledge protects against — and what it doesn't

It protects against

• Provider-side insider threats — the operator's staff cannot read your data even with full system access.
• Compelled disclosure to the operator — a subpoena yields metadata and ciphertext only.
• Operator data breaches — exfiltrated databases contain ciphertext fragments and wrapped keys, not plaintext.
• Operator-side machine learning — the platform cannot train models on your content because it cannot read it.

It does not protect against

• Compromise of the customer's own device once they've authenticated and decrypted content locally.
• A customer voluntarily sharing decrypted content with another party.
• Loss of the password AND the recovery key AND any duress-mode credentials (the architecture cannot recover from total credential loss — this is a feature, not a bug).

These are properties of the threat model: zero-knowledge moves the trust boundary to the customer device. Securing that endpoint is then the customer's responsibility — which most organisations are already well-equipped to do through standard endpoint security practices.

Zero-knowledge ≠ end-to-end encryption

The two terms are often conflated. End-to-end encryption (E2EE) means data is encrypted at the sender's device and decrypted only at the recipient's device. Zero-knowledge means the storage operator has no key material in unwrapped form and no API path that returns plaintext. SkyeConnex is both — but they're distinct guarantees. A messaging app can be E2EE without being zero-knowledge (if the operator holds the key escrow). A storage system can be zero-knowledge for a single user without being E2EE in the sharing sense (if shares are wrapped server-side).

How to evaluate a zero-knowledge claim

Vendors often claim zero-knowledge without delivering it. Three diagnostic questions:

1. Where is the key derived? If anywhere on the vendor's infrastructure, it's not zero-knowledge.
2. Does any server-side API return plaintext? If yes — even under "authorised" conditions — it's not zero-knowledge.
3. Can the vendor decrypt under compulsion? If yes (even reluctantly), the architecture is not zero-knowledge.

SkyeConnex answers: client device only, no, and no — by construction. The capability to decrypt does not exist on our infrastructure.